Plan7¶

High-level interface to the Plan7 data model.

Plan7 is the model architecture used by HMMER since HMMER2.

Details about the Plan 7 architecture in the HMMER documentation.

Hidden Markov Model¶

HMM¶

class pyhmmer.plan7.HMM

A data structure storing the Plan7 Hidden Markov Model.

alphabet

The alphabet of the model.

Type

Alphabet

Changed in version 0.4.6: Added the evalue_parameters and cutoffs attributes.

__init__(M, alphabet)

Create a new HMM from scratch.

Parameters
copy()

Return a copy of the HMM with the exact same configuration.

New in version 0.3.0.

match_occupancy()

Calculate the match occupancy for each match state.

Returns

VectorF – A vector of size $$M+1$$ containing the probability that each match state is used in a sample glocal path through the model.

New in version 0.4.10.

mean_match_entropy()

Compute the mean entropy per HMM match state.

The mean entropy per match state emission distribution is defined as:

$- \frac{1}{M} \sum_{k=1}^{M} \sum_x p_k(x) \log_2 p_k(x)$

where $$p_k(x)$$ is the emission probability for symbol $$x$$ from match state $$k$$.

Returns

float – The mean entropy, in bits.

Example

>>> thioesterase.mean_match_entropy()
3.0425...


New in version 0.4.10.

mean_match_information(background)

Compute the mean information content of the HMM match states.

The mean information content per match state emission distribution is defined as:

$\frac{1}{M} \sum_{k=1}^{M} \left[ \sum_x f(x) \log_2 f(x) - \sum_x p_k(x) \log_2 p_k(x) \right]$

in bits, where $$p_k(x)$$ is the emission probability for symbol $$x$$ from match state $$k$$, and $$f(x)$$ is the background emission probability for $$x$$ from the null model.

Parameters

background (Background) – The null model from which to get the background emission probabilities.

Returns

float – The mean information content, in bits.

Example

>>> background = plan7.Background(easel.Alphabet.amino())
>>> thioesterase.mean_match_information(background)
1.1330...


New in version 0.4.10.

mean_match_relative_entropy(background)

Compute the mean relative entropy per HMM match state.

The mean relative entropy per match state emission distribution is defined as:

$\frac{1}{M} \sum_{k=1}^{M} \sum_x { p_k(x) \log_2 \frac{p_k(x)}{f(x)} }$

in bits, where $$p_k(x)$$ is the emission probability for symbol $$x$$ from match state $$k$$, and $$f(x)$$ is the background emission probability for $$x$$ from the null model.

Parameters

background (Background) – The null model from which to get the background emission probabilities.

Returns

float – The mean relative entropy, in bits.

Example

>>> background = plan7.Background(easel.Alphabet.amino())
>>> thioesterase.mean_match_relative_entropy(background)
1.1201...


New in version 0.4.10.

renormalize()

Renormalize all parameter vectors (emissions and transitions).

New in version 0.4.0.

scale(scale, exponential=False)

Rescale counts by a factor in a model containing counts.

This method only affects core probability model emissions and transitions.

Parameters
• scale (float) – The scaling factor to use ($$1.0$$ for no scaling). Often computed using the ratio of effective sequences ($$\frac{n_{eff}}{n_{seq}}$$)

• exponential (bool) – When set to True, use scale as an exponential factor ($$C_i = C_i^s$$) instead of a multiplicative factor ($$C_i = C_i \times s$$), resulting in a non-uniform scaling across columns. This can be useful when some heavily fragmented sequences are used to reconstruct a family MSA.

New in version 0.4.0.

set_composition()

Calculate and set the model composition.

New in version 0.4.0.

write(fh, binary=False)

Write the HMM to a file handle.

Parameters
• fh (io.IOBase) – A Python file handle, opened in binary mode (this must be the case even with binary=False, since the C code will emit bytes in either case).

• binary (bool) – Pass False to emit the file in ASCII mode using the latest supported HMMER format, or True to use the binary HMMER3 format.

zero()

Set all parameters to zero, including model composition.

M

The length of the model (i.e. the number of nodes).

Type

int

accession

The accession of the HMM, if any.

Type
checksum

The 32-bit checksum of the HMM, if any.

The checksum if calculated from the alignment the HMM was created from, and was introduced in more recent HMM formats. This means some HMM objects may have a non-None checksum.

New in version 0.2.1.

Changed in version 0.3.1: Returns None if the HMM flag for the checksum is not set.

Type
command_line

The command line that built the model.

For HMMs created with Builder, this defaults to sys.argv. It can however be set to any string, including multiline to show successive commands.

Example

>>> print(thioesterase.command_line)
hmmbuild Thioesterase.hmm Thioesterase.fa
hmmcalibrate Thioesterase.hmm


New in version 0.3.1.

Type
composition

The model composition.

May not be available for freshly-created HMMs. To get the mean residue composition emitted by the model, the set_composition method must be called to compute the composition from occupancy.

Note

Although the allocated buffer in the P7_HMM object is always the same dimension so that it can store the largest possible alphabet, we only expose the relevant residues. This means that the vector will be of size alphabet.K:

>>> dna = easel.Alphabet.dna()  # dna.K=4
>>> hmm = plan7.HMM(100, dna)
>>> hmm.set_composition()
>>> len(hmm.composition)
4


New in version 0.4.0.

Type
consensus

The consensus residue line of the HMM, if set.

New in version 0.3.0.

Type
consensus_accessibility

The consensus accessibility of the HMM, if any.

New in version 0.3.1.

Type
consensus_structure

The consensus structure of the HMM, if any.

New in version 0.3.1.

Type
creation_time

The creation time of the HMM, if any.

Example

Get the creation time for any HMM:

>>> thioesterase.creation_time
datetime.datetime(2008, 11, 25, 17, 28, 32)


Set the creation time manually to a different date and time:

>>> ctime = datetime.datetime(2021, 8, 23, 23, 59, 19)
>>> thioesterase.creation_time = ctime
>>> thioesterase.creation_time
datetime.datetime(2021, 8, 23, 23, 59, 19)


Danger

Internally, libhmmer always uses asctime to generate a timestamp for the HMMs, so this property assumes that every creation time field can be parsed into a datetime.datetime object using the "%a %b %d %H:%M:%S %Y" format.

New in version 0.4.6.

Type
cutoffs

The bitscore cutoffs for this HMM.

Type

Cutoffs

description

The description of the HMM, if any.

Type
evalue_parameters

The e-value parameters for this HMM.

Type

EvalueParameters

insert_emissions

The insert emissions of the model.

The property exposes a matrix of shape $$(M+1, K)$$, with one row per node and one column per alphabet symbol.

Caution

If editing this matrix manually, note that rows must contain valid probabilities for the HMM to be valid: each row must contains values between 0 and 1, and sum to 1.

New in version 0.3.1.

Changed in version 0.4.0: This property is now a MatrixF, and stores row 0.

Type

MatrixF

match_emissions

The match emissions of the model.

The property exposes a matrix of shape $$(M+1, K)$$, with one row per node and one column per alphabet symbol.

Note

Since the first row corresponds to the entry probabilities, the emissions are unused. By convention, it should still contain valid probabilities, so it will always be set as follow with 1 probability for the first symbol, and 0 for the rest:

>>> hmm = HMM(100, alphabet=easel.Alphabet.dna())
>>> hmm.match_emissions[0]
pyhmmer.easel.VectorF([1.0, 0.0, 0.0, 0.0])


Caution

If editing this matrix manually, note that rows must contain valid probabilities for the HMM to be valid: each row must contains values between 0 and 1, and sum to 1.

New in version 0.3.1.

Changed in version 0.4.0: This property is now a MatrixF, and stores row 0.

Type

MatrixF

The model mask line from the alignment, if any.

New in version 0.3.1.

Type
name

The name of the HMM, if any.

Type
nseq

The number of training sequences used, if any.

If the HMM was created from a multiple sequence alignment, this corresponds to the number of sequences in the MSA.

Example

>>> thioesterase.nseq
278


New in version 0.3.1.

Type
nseq_effective

The number of effective sequences used, if any.

New in version 0.3.1.

Type
reference

The reference line from the alignment, if any.

This is relevant if the HMM was built from a multiple sequence alignment (e.g. by Builder.build_msa, or by an external hmmbuild pipeline run).

New in version 0.3.1.

Type
transition_probabilities

The transition probabilities of the model.

The property exposes a matrix of shape $$(M+1, 7)$$, with one row per node (plus one extra row for the entry probabilities), and one column per transition.

Columns indices correspond to the following transitions weights:

• 0 for $$M_n \to M_{n+1}$$

• 1 for $$M_n \to I_{n+1}$$

• 2 for $$M_n \to D_{n+1}$$

• 3 for $$I_n \to M_{n+1}$$

• 4 for $$I_n \to I_{n+1}$$

• 5 for $$D_n \to M_{n+1}$$

• 6 for $$D_n \to D_{n+1}$$

Example

>>> t = thioesterase.transition_probabilities
>>> t[0, 5]  # TDM, 1 by convention
1.0


Caution

If editing this matrix manually, note that some invariants need to hold for the HMM to be valid: $$I_n$$, $$M_n$$ and $$D_n$$ transition probabilities should only contain probabilities between 0 and 1, and sum to 1:

>>> t = thioesterase.transition_probabilities
>>> t[50, 0] + t[50, 1] + t[50, 2]  # M_n probabilities
1.000...
>>> t[50, 3] + t[50, 4]  # I_n probabilities
1.000...
>>> t[50, 5] + t[50, 6]  # D_n probabilties
1.000...


New in version 0.3.1.

Changed in version 0.4.0: This property is now a MatrixF.

Type

MatrixF

HMM File¶

class pyhmmer.plan7.HMMFile

A wrapper around a file (or database), storing serialized HMMs.

__init__(file, db=True)

Create a new HMM reader from the given file.

Parameters
close()

Close the HMM file and free resources.

This method has no effect if the file is already closed. It is called automatically if the HMMFile was used in a context:

>>> with HMMFile("tests/data/hmms/bin/PKSI-AT.h3m") as hmm_file:
...     hmm = hmm_file.read()

is_pressed()

Check whether the HMM file is a pressed HMM database.

A pressed database is an HMMER format to store optimized profiles in several files on the disk. It can be used to reduce the time needed to process sequences by cutting down the time needed to convert from an HMM to an OptimizedProfile.

Example

>>> HMMFile("tests/data/hmms/txt/PKSI-AT.hmm").is_pressed()
False
>>> HMMFile("tests/data/hmms/bin/PKSI-AT.h3m").is_pressed()
False
>>> HMMFile("tests/data/hmms/db/PKSI-AT.hmm").is_pressed()
True


New in version 0.4.11.

optimized_profiles()

Get an iterator over the OptimizedProfile in the HMM database.

Returns

HMMPressedFile – An iterator over the optimized profiles in a pressed HMM database.

New in version 0.4.11.

Read the next HMM from the file.

Returns

HMM – The next HMM in the file, or None if all HMMs were read from the file already.

Raises
• ValueError – When attempting to read a HMM from a closed file, or when the file could not be parsed.

• AllocationError – When memory for the HMM could not be allocated successfully.

New in version 0.4.11.

closed

Whether the HMMFile is closed or not.

Type

bool

Profile¶

Profile¶

class pyhmmer.plan7.Profile

A Plan7 search profile.

alphabet

The alphabet for which this profile was configured.

Type

Alphabet

Changed in version 0.4.6: Added the evalue_parameters and cutoffs attributes.

__init__(M, alphabet)

Create a new profile for the given alphabet.

Parameters
clear()

Clear internal buffers to reuse the profile without reallocation.

configure(hmm, background, L, multihit=True, local=True)

Configure a search profile using the given models.

Parameters
copy()

Return a copy of the profile with the exact same configuration.

is_local()

Return whether or not the profile is in a local alignment mode.

is_multihit()

Returns whether or not the profile is in a multihit alignment mode.

optimized()

Convert the profile to a platform-specific optimized profile.

Returns

OptimizedProfile – The platform-specific optimized profile built using the configuration of this profile.

L

The current configured target sequence length.

Type

int

M

The length of the profile (i.e. the number of nodes).

New in version 0.3.0.

Type

int

accession

The accession of the profile, if any.

New in version 0.3.0.

Type
consensus

The consensus residue line of the profile, if set.

New in version 0.4.1.

Type
consensus_structure

The consensus structure of the profile, if any.

New in version 0.4.1.

Type
cutoffs

The bitscore cutoffs for this profile, if any.

Type

Cutoffs

description

The description of the profile, if any.

New in version 0.3.0.

Type
evalue_parameters

The e-value parameters for this profile.

Type

EvalueParameters

name

The name of the profile, if any.

New in version 0.3.0.

Type
offsets

The disk offsets for this profile.

Type

Offsets

OptimizedProfile¶

class pyhmmer.plan7.OptimizedProfile

An optimized profile that uses platform-specific instructions.

Optimized profiles store the match emissions and transition probabilities for a profile HMM so that they can be loaded in the SIMD code. Typically, matrices use aligned storage so that they can loaded efficiently, and are striped à la Farrar to compute pairwise scores for each sequence residue and profile node.

alphabet

The alphabet for which this optimized profile is configured.

Type

Alphabet

References

• Farrar, Michael. Striped Smith–Waterman Speeds Database Searches Six Times over Other SIMD Implementations. Bioinformatics 23, no. 2 (15 January 2007): 156–61. doi:10.1093/bioinformatics/btl582.

__init__(M, alphabet)

Create a new optimized profile from scratch.

Once allocated, you must call the convert method with a Profile object. It’s actually easier to use Profile.optimized method to obtained a configured OptimizedProfile directly, unless you’re explicitly trying to recycle memory.

Parameters
convert(profile)

Store the given profile into self as a platform-specific profile.

Use this method to obtained an optimized profile from a Profile while recycling the internal vector buffers.

The Profile.optimized method, which allows getting an OptimizedProfile directly from a profile without having to allocate first.

copy()

Create an exact copy of the optimized profile.

is_local()

Return whether or not the profile is in a local alignment mode.

ssv_filter(seq)

Compute the SSV filter score for the given sequence.

Parameters

seq (DigitalSequence) – The sequence in digital format for which to compute the SSV filter score.

Returns

float or None – The SSV filter score for the sequence.

Note

• math.inf may be returned if an overflow occurs that will also occur in the MSV filter. This is the case whenever $$\text{base} - \text{tjb} - \text{tbm} \ge 128$$

• None may be returned if the MSV filter score needs to be recomputed (because it may not overflow even though the SSV filter did).

Raises

AlphabetMismatch – When the alphabet of the sequence does not correspond to the profile alphabet.

Caution

This method is not available on the PowerPC platform (calling it will raise a NotImplementedError).

New in version 0.4.0.

write(fh_filter, fh_profile)

Write an optimized profile to two separate files.

HMMER implements an acceleration pipeline using several scoring algorithms. Parameters for MSV (the Multi ungapped Segment Viterbi) are saved independently to the fh_filter handle, while the rest of the profile is saved to fh_profile.

L

The currently configured target sequence length.

New in version 0.4.0.

Type

int

M

The number of nodes in the model.

New in version 0.4.0.

Type

int

accession

The accession of the profile, if any.

New in version 0.4.11.

Type
bias

The positive bias to emission scores.

New in version 0.4.0.

Type

int

consensus

The consensus residue line of the profile, if any.

New in version 0.4.11.

Type
consensus_structure

The consensus structure of the profile, if any.

New in version 0.4.11.

Type
cutoffs

The bitscore cutoffs for this profile, if any.

Type

Cutoffs

description

The description of the profile, if any.

New in version 0.4.11.

Type
evalue_parameters

The e-value parameters for this profile.

Type

EvalueParameters

name

The name of the profile, if any.

New in version 0.4.11.

Type
offsets

The disk offsets for this optimized profile.

Type

Offsets

rbv

The match scores for the MSV filter.

Type

MatrixU8

sbv

The match scores for the SSV filter.

New in version 0.4.0.

Type

MatrixU8

tbm

The constant cost for a $$B \to M_k$$ transition.

New in version 0.4.0.

Type

int

tec

The constant cost for a $$E \to C$$ transition.

New in version 0.4.0.

Type

int

tjb

The constant cost for a $$NJC$$ move.

New in version 0.4.0.

Type

int

Background¶

class pyhmmer.plan7.Background

The null background model of HMMER.

alphabet

The alphabet of the backgound model.

Type

Alphabet

uniform

Whether or not the null model has been created with uniform frequencies.

Type

bool

residue_frequencies

The null1 background residue frequencies.

Type

VectorF

Changed in version 0.4.0: Added the residue_frequencies attribute.

__init__(alphabet, uniform=False)

Create a new background model for the given alphabet.

Parameters
copy()

Create a copy of the null model with the same parameters.

L

The mean of the null model length distribution, in residues.

Type

int

omega

The prior on null2/null3.

New in version 0.4.0.

Type

float

transition_probability

The null1 transition probability ($$\frac{L}{L+1}$$).

New in version 0.4.0.

Type

float

Pipelines¶

Pipeline¶

class pyhmmer.plan7.Pipeline

An HMMER3 accelerated sequence/profile comparison pipeline.

The Plan7 pipeline handles the platform-accelerated comparison of sequences to profile HMMs. It performs either a search (comparing a single query profile to a target sequence database) or a scan (comparing a single query sequence to a target profile database). The two methods are yielding equivalent results: if you have a collection of $$M$$ sequences and $$N$$ HMMs to compare, doing a search or a scan should give the same hits.

The main reason for which you should choose search or scan is the relative size of the sequences and HMMs databases. In the original HMMER3 code, the memory was managed in a way that you never had to load the entirety of the target sequences in memory. In PyHMMER, the search methods will require that you have the entirety of target sequences loaded in memory, which may not be feasible if you have too many sequences.

alphabet

The alphabet for which the pipeline is configured.

Type

Alphabet

background

The null background model to use to compute scores.

Type

Background

randomness

The random number generator being used by the pipeline.

Type

Randomness

Changed in version 0.4.2: Added the randomness attribute.

__init__(alphabet, background=None, *, bias_filter=True, null2=True, seed=42, Z=None, domZ=None, F1=0.02, F2=0.001, F3=1e-05, E=10.0, T=None, domE=10.0, domT=None, incE=0.01, incT=None, incdomE=0.01, incdomT=None, bit_cutoffs=None)

Instantiate and configure a new accelerated comparison pipeline.

Parameters
• alphabet (Alphabet) – The biological alphabet the of the HMMs and sequences that are going to be compared. Used to build the background model.

• background (Background, optional) – The background model to use with the pipeline, or None to create and use a default one. The pipeline needs ownership of the background model, so any background model passed there will be copied.

Keyword Arguments

Hint

In order to run the pipeline in slow/max mode, disable the bias filter, and set the three filtering parameters to $$1.0$$:

>>> dna = easel.Alphabet.dna()
>>> max_pli = Pipeline(dna, bias_filter=False, F1=1.0, F2=1.0, F3=1.0)


Changed in version 0.4.6: Added keywords arguments to set the E-value thresholds.

arguments()

Generate an argv-like list with pipeline options as CLI flags.

Example

>>> alphabet = easel.Alphabet.amino()
>>> plan7.Pipeline(alphabet).arguments()
[]
>>> plan7.Pipeline(alphabet, F1=0.01).arguments()
['--F1', '0.01']


New in version 0.6.0.

clear()

Reset the pipeline configuration to its default state.

iterate_hmm(query, sequences, builder=None, select_hits=None)

Run the pipeline to find homologous sequences to a query HMM.

Parameters
• query (HMM) – The sequence object to use to query the sequence database.

• sequences (collection of DigitalSequence) – The sequences to query.

• builder (Builder, optional) – A HMM builder to use to convert the query and subsequent alignments to a HMM. If None is given, this method will create one with the default parameters.

• select_hits (callable, optional) – A function or callable object for manually selecting hits during each iteration. It should take a single TopHits argument and change the inclusion of individual hits with the Hit.include and Hit.drop methods.

Returns

IterativeSearch – An iterator object yielding the hits, sequence alignment, and HMM for each iteration.

Raises

AlphabetMismatch – When the alphabet of the current pipeline does not match the alphabet of the given query or database sequences.

Hint

This method corresponds to running jackhmmer with the query sequence against the sequences database.

Caution

Default values used for jackhmmer do not correspond to the default parameters used for creating a pipeline in the other cases. To have truly identical results to the jackhmmer results in default mode, create the Pipeline object with incE=0.001 and incdomE=0.001.

The iterate_seq, which does the same operation with a query sequence instead of a query HMM, and contains more details and examples.

iterate_seq(query, sequences, builder=None, select_hits=None)

Run the pipeline to find homologous sequences to a query sequence.

This method implements an iterative search over a database to find all sequences homologous to a query sequence. It is very sensitive to the pipeline inclusion thresholds (incE and incdomE).

Since this method returns an iterator, the local results of each iteration will be available for inspection before starting the next one. The select_hits callback in particular can be used for manually including/excluding hits in each iteration, which is not supported in the original jackhmmer, but available on the HMMER web client.

Parameters
• query (DigitalSequence) – The sequence object to use to query the sequence database.

• sequences (collection of DigitalSequence) – The sequences to query.

• builder (Builder, optional) – A HMM builder to use to convert the query and subsequent alignments to a HMM. If None is given, this method will create one with the default parameters.

• select_hits (callable, optional) – A function or callable object for manually selecting hits during each iteration. It should take a single TopHits argument and change the inclusion of individual hits with the Hit.include and Hit.drop methods.

Returns

IterativeSearch – An iterator object yielding the hits, sequence alignment, and HMM for each iteration.

Raises

AlphabetMismatch – When the alphabet of the current pipeline does not match the alphabet of the given query or database sequences.

Hint

This method corresponds to running jackhmmer with the query sequence against the sequences database.

Caution

Default values used for jackhmmer do not correspond to the default parameters used for creating a pipeline in the other cases. To have truly identical results to the jackhmmer results in default mode, create the Pipeline object with incE=0.001 and incdomE=0.001.

Example

Starting from a pipeline and a query sequence, let’s first obtain the iterator over the successive results:

>>> abc = easel.Alphabet.amino()
>>> pli = plan7.Pipeline(abc, incE=1e-3, incdomE=1e-3)
>>> iterator = pli.iterate_seq(reductase, proteins)


Once this is ready, we can keep iterating until we converge:

>>> converged = False
>>> while not converged:
...     _, hits, _, converged, _ = next(iterator)
...     print(f"Hits: {len(hits)}  Converged: {converged}")
Hits: 1  Converged: False
Hits: 2  Converged: False
Hits: 2  Converged: True


To prevent diverging searches from running infinitely, you could wrap the search in a for loop instead, using a number of maximum iterations as the upper boundary:

>>> iterator = pli.iterate_seq(reductase, proteins)
>>> max_iterations = 10
>>> for n in range(max_iterations):
...     iteration = next(iterator)
...     if iteration.converged:
...         break


New in version 0.6.0.

scan_seq(query, hmms)

Run the pipeline using a query sequence against a profile database.

Parameters
Returns

TopHits – the hits found in the profile database.

Raises

AlphabetMismatch – When the alphabet of the current pipeline does not match the alphabet of the given query or profile.

Caution

In the current version, this method is not optimized to use the pressed database, even if it exists. This will cause the MSV and SSV filters to be rebuilt at each iteration, which could be slow. Consider at least pre-fetching the HMM database if calling this method several times in a row.

Hint

This method corresponds to running hmmscan with the query sequence against the hmms database.

New in version 0.4.0.

search_hmm(query, sequences)

Run the pipeline using a query HMM against a sequence database.

Parameters
Returns

TopHits – the hits found in the sequence database.

Raises
• ValueError – When the pipeline is configured to use model-specific reporting thresholds but the HMM query doesn’t have the right cutoffs available.

• AlphabetMismatch – When the alphabet of the current pipeline does not match the alphabet of the given HMM.

Hint

This method corresponds to running hmmsearch with the query HMM against the sequences database.

New in version 0.2.0.

Changed in version 0.4.9: Query can now be a Profile or an OptimizedProfile.

search_msa(query, sequences, builder=None)

Run the pipeline using a query alignment against a sequence database.

Parameters
Returns

TopHits – the hits found in the sequence database.

Raises

Hint

This method corresponds to running phmmer with the query multiple sequence alignment against the sequences database.

Caution

Internally, this method will create a new HMM from the query MSA using the Builder.build_msa method. HMMER requires that every HMM has a name, so the Builder will attempt to use the name of the query MSA to name the HMM. Passing an MSA without a name will result in an error.

New in version 0.3.0.

search_seq(query, sequences, builder=None)

Run the pipeline using a query sequence against a sequence database.

Parameters
Returns

TopHits – the hits found in the sequence database.

Raises

Hint

This method corresponds to running phmmer with the query sequence against the sequences database.

New in version 0.2.0.

E

The per-target E-value threshold for reporting a hit.

New in version 0.4.6.

Type

float

F1

The MSV filter threshold.

New in version 0.4.1.

Type

float

F2

The Viterbi filter threshold.

New in version 0.4.1.

Type

float

F3

The uncorrected Forward filter threshold.

New in version 0.4.1.

Type

float

T

The per-target score threshold for reporting a hit.

If set to a non-None value, this threshold takes precedence over the per-target E-value threshold (Pipeline.E).

New in version 0.4.8.

Type
Z

The number of effective targets searched.

It is used to compute the independent e-value for each domain, and for an entire hit. If None, the parameter number will be set automatically after all the comparisons have been done. Otherwise, it can be set to an arbitrary number.

Type
bias_filter

Whether or not to enable the biased comp HMM filter.

New in version 0.4.1.

Type

bool

bit_cutoffs

The model-specific thresholding option, if any.

New in version 0.4.6.

Type
domE

The per-domain E-value threshold for reporting a hit.

New in version 0.4.6.

Type

float

domT

The per-domain score threshold for reporting a hit.

If set to a non-None value, this threshold takes precedence over the per-domain E-value threshold (Pipeline.domE).

New in version 0.4.8.

Type
domZ

The number of significant targets.

It is used to compute the conditional e-value for each domain. If None, the parameter number will be set automatically after all the comparisons have been done, and all hits have been thresholded. Otherwise, it can be set to an arbitrary number.

Type
incE

The per-target E-value threshold for including a hit.

New in version 0.4.6.

Type

float

incT

The per-target score threshold for including a hit.

If set to a non-None value, this threshold takes precedence over the per-target E-value inclusion threshold (Pipeline.incE).

New in version 0.4.8.

Type
incdomE

The per-domain E-value threshold for including a hit.

New in version 0.4.6.

Type

float

incdomT

The per-domain score threshold for including a hit.

If set to a non-None value, this threshold takes precedence over the per-domain E-value inclusion threshold (Pipeline.incdomE).

New in version 0.4.8.

Type
null2

Whether or not to enable the null2 score correction.

New in version 0.4.1.

Type

bool

seed

The seed given at pipeline initialization.

Setting this attribute to a different value will cause the random number generator to be reseeded immediately.

New in version 0.2.0.

Changed in version 0.4.2: Avoid shadowing initial null seeds given on pipeline initialization.

Type

int

LongTargetsPipeline¶

class pyhmmer.plan7.LongTargetsPipeline(Pipeline)

An HMMER3 pipeline tuned for long targets.

The default HMMER3 pipeline is configured not to accept target sequences longer than 100,000 residues. Although there is no strong limitation for this threshold, comparing a sequence of $$L$$ residues to a profile with $$M$$ nodes requires the allocation of a $$L imes M$$ dynamic programming matrix.

For sequences too long, it’s actually more efficient memory-wise to use a sliding window to match the profile to the sequence. The usual comparison pipeline is then used to perform the comparison on each window, and results are merged once the entire sequence is done being processed. The context size $$C$$ is large enough to accommodate for the entire profile, so that there is no risk of missing a hit in the overlaps between windows. The window size $$W$$ can be changed with the block_length argument when instantiating a new LongTargetsPipeline object.

New in version 0.4.9.

__init__(alphabet, background=None, *, F1=0.02, F2=0.003, F3=3e-05, strand=None, B1=100, B2=240, B3=1000, block_length=262144, **kwargs)

Instantiate and configure a new long targets pipeline.

Parameters
• alphabet (Alphabet) – The biological alphabet the of the HMMs and sequences that are going to be compared. Used to build the background model. A nucleotide alphabet is expected.

• background (Background, optional) – The background model to use with the pipeline, or None to create and use a default one. The pipeline needs ownership of the background model, so any background model passed there will be copied.

Keyword Arguments
• strand (str, optional) – The strand to use when processing nucleotide sequences. Use "watson" to use only the coding strand, "crick" to use only the reverse strand, or leave as None to process both strands.

• B1 (int) – The window length to use for the biased-composition modifier of the MSV filter.

• B2 (int) – The window length to use for the biased-composition modifier of the Viterbi filter.

• B3 (int) – The window length to use for the biased-composition modifier of the Forward filter.

• block_length (int) – The number of residues to use as the window size $$W$$ when reading blocks from the long target sequences.

• **kwargs – Any additional parameter will be passed to the Pipeline constructor.

arguments()

Generate an argv-like list with pipeline options as CLI flags.

Example

>>> alphabet = easel.Alphabet.dna()
>>> plan7.LongTargetsPipeline(alphabet).arguments()
[]
>>> plan7.LongTargetsPipeline(alphabet, B1=200).arguments()
['--B1', '200']


New in version 0.6.0.

scan_seq()

Run the pipeline using a query sequence against a profile database.

Parameters
Returns

TopHits – the hits found in the profile database.

Raises

AlphabetMismatch – When the alphabet of the current pipeline does not match the alphabet of the given query or profile.

Caution

In the current version, this method is not optimized to use the pressed database, even if it exists. This will cause the MSV and SSV filters to be rebuilt at each iteration, which could be slow. Consider at least pre-fetching the HMM database if calling this method several times in a row.

Hint

This method corresponds to running hmmscan with the query sequence against the hmms database.

New in version 0.4.0.

search_hmm(query, sequences)

Run the pipeline using a query HMM against a sequence database.

Parameters
Returns

TopHits – the hits found in the sequence database.

Raises
• ValueError – When the pipeline is configured to use model-specific reporting thresholds but the HMM query doesn’t have the right cutoffs available.

• AlphabetMismatch – When the alphabet of the current pipeline does not match the alphabet of the given HMM.

Hint

This method corresponds to running nhmmer with the query HMM against the sequences database.

search_msa(query, sequences, builder=None)

Run the pipeline using a query alignment against a sequence database.

Parameters
Returns

TopHits – the hits found in the sequence database.

Raises

AlphabetMismatch – When the alphabet of the current pipeline does not match the alphabet of the given query.

Hint

This method corresponds to running nhmmer with the query multiple sequence alignment against the sequences database.

search_seq(query, sequences, builder=None)

Run the pipeline using a query sequence against a sequence database.

Parameters
Returns

TopHits – the hits found in the sequence database.

Raises

AlphabetMismatch – When the alphabet of the current pipeline does not match the alphabet of the given query.

Hint

This method corresponds to running nhmmer with the query sequence against the sequences database.

B1

The window length for biased-composition modifier of the MSV.

Type

int

B2

The window length for biased-composition modifier of the Viterbi.

Type

int

B3

The window length for biased-composition modifier of the Forward.

Type

int

strand

The strand to process, or None for both.

Type

Builder¶

class pyhmmer.plan7.Builder

A factory for constructing new HMMs from raw sequences.

alphabet

The alphabet the builder is configured to use to convert sequences to HMMs.

Type

Alphabet

randomness

The random number generator being used by the builder.

Type

Randomness

score_matrix

The name of the substitution matrix used to build HMMs for single sequences.

Type

str

popen

The gap open probability to use when building HMMs from single sequences.

Type

float

pextend

The gap extend probability to use when building HMMs from single sequences.

Type

float

New in version 0.2.0.

Changed in version 0.4.2: Added the randomness attribute.

__init__(alphabet, *, architecture='fast', weighting='pb', effective_number='entropy', prior_scheme='alpha', symfrac=0.5, fragthresh=0.5, wid=0.62, esigma=45.0, eid=0.62, EmL=200, EmN=200, EvL=200, EvN=200, EfL=100, EfN=200, Eft=0.04, seed=42, ere=None, popen=None, pextend=None, window_length=None, window_beta=None)

Create a new sequence builder with the given configuration.

Parameters

alphabet (Alphabet) – The alphabet the builder expects the sequences to be in.

Keyword Arguments
• architecture (str) – The algorithm to use to determine the model architecture, either "fast" (the default), or "hand".

• weighting (str) – The algorithm to use for relative sequence weighting, either "pb" (the default), "gsc", "blosum", "none", or "given".

• effective_number (str, int, or float) – The algorithm to use to determine the effective sequence number, either "entropy" (the default), "exp", "clust", "none". A number can also be given directly to set the effective sequence number manually.

• prior_scheme (str) – The choice of mixture Dirichlet prior when parameterizing from counts, either "laplace" or "alphabet" (the default).

• symfrac (float) – The residue occurrence threshold for fast architecture determination.

• fragthresh (float) – A threshold such that a sequence is called a fragment when $$L \le fragthresh imes alen$$.

• wid (double) – The percent identity threshold for BLOSUM relative weighting.

• esigma (float) – The minimum total relative entropy parameter for effective number entropy weights.

• eid (float) – The percent identity threshold for effective number clustering.

• EmL (int) – The length of sequences generated for MSV fitting.

• EmN (int) – The number of sequences generated for MSV fitting.

• EvL (int) – The lenght of sequences generated for Viterbi fitting.

• EvN (int) – The number of sequences generated for Viterbi fitting.

• EfL (int) – The lenght of sequences generated for Forward fitting.

• EfN (int) – The number of sequences generated for Forward fitting.

• Eft (float) – The tail mass used for Forward fitting.

• seed (int) – The seed to use to initialize the internal random number generator. If 0 is given, an arbitrary seed will be chosen based on the current time.

• ere (double, optional) – The relative entropy target for effective number weighting, or None.

• popen (float, optional) – The gap open probability to use when building HMMs from single sequences. The default value depends on the alphabet: 0.02 for proteins, 0.03125 for nucleotides.

• pextend (float, optional) – The gap extend probability to use when building HMMs from single sequences. Default depends on the alphabet: 0.4 for proteins, 0.75 for nucleotides.

• score_matrix (str, optional) – The name of the score matrix to use when building HMMs from single sequences. The only allowed value for nucleotide alphabets is DNA1. For proteins, PAM30, PAM70, PAM120, PAM240, BLOSUM45, BLOSUM50, BLOSUM62 (the default), BLOSUM80 or BLOSUM90 can be used.

• window_length (float, optional) – The window length for nucleotide sequences, essentially the max expected hit length. If given, takes precedence over window_beta.

• window_beta (float, optional) – The tail mass at which window length is determined for nucleotide sequences.

build(sequence, background)

Build a new HMM from sequence using the builder configuration.

Parameters
• sequence (DigitalSequence) – A single biological sequence in digital mode to build a HMM with.

• background (pyhmmer.plan7.background) – The background model to use to create the HMM.

Returns

(HMM, Profile, OptimizedProfile) – A tuple containing the new HMM as well as profiles to be used directly in a Pipeline.

Raises
• AlphabetMismatch – When either sequence or background have the wrong alphabet for this builder.

• ValueError – When the HMM cannot be created successfully from the input; the error message contains more details.

Hint

The score matrix and the gap probabilities used here can be set when initializing the builder, or changed by setting a new value to the right attribute:

>>> alphabet = easel.Alphabet.amino()
>>> background = plan7.Background(alphabet)
>>> builder = plan7.Builder(alphabet, popen=0.05)
>>> builder.score_matrix = "BLOSUM90"
>>> hmm, _, _ = builder.build(proteins[0], background)


Changed in version 0.4.6: Sets the HMM.creation_time attribute with the current time.

Changed in version 0.4.10: The score system is now cached between calls to Builder.build.

build_msa(msa, background)

Build a new HMM from msa using the builder configuration.

Parameters
• msa (DigitalMSA) – A multiple sequence alignment in digital mode to build a HMM with.

• background (pyhmmer.plan7.background) – The background model to use to create the HMM.

Returns

(HMM, Profile, OptimizedProfile) – A tuple containing the new HMM as well as profiles to be used directly in a Pipeline.

Raises
• AlphabetMismatch – When either msa or background have the wrong alphabet for this builder.

• ValueError – When the HMM cannot be created successfully from the input; the error message contains more details.

Caution

HMMER requires that every HMM has a name, so the Builder will attempt to use the name of the sequence to name the HMM. Passing an MSA without a name will result in an error:

>>> alphabet = easel.Alphabet.amino()
>>> msa = easel.TextMSA(sequences=[
...   easel.TextSequence(name=b"ustiA", sequence="YAIG"),
...   easel.TextSequence(name=b"ustiB", sequence="YVIG")
... ])
>>> builder = plan7.Builder(alphabet)
>>> background = plan7.Background(alphabet)
>>> builder.build_msa(msa.digitize(alphabet), background)
Traceback (most recent call last):
...
ValueError: Could not build HMM: Unable to name the HMM.


New in version 0.3.0.

Changed in version 0.4.6: Sets the HMM.creation_time attribute with the current time.

copy()

Create a duplicate Builder instance with the same arguments.

seed

The seed given at builder initialization.

Setting this attribute to a different value will cause the internal random number generator to be reseeded immediately.

Changed in version 0.4.2: Avoid shadowing initial null seeds given on builder initialization.

Type

int

window_beta

The tail mass at which window length is determined.

Type

float

window_length

The window length for nucleotide sequences.

Type

Results¶

TopHits¶

class pyhmmer.plan7.TopHits

An immutable ranked list of top-scoring hits.

TopHits are thresholded using the parameters from the pipeline, and are sorted by key when you obtain them from a Pipeline instance:

>>> abc = thioesterase.alphabet
>>> hits = Pipeline(abc).search_hmm(thioesterase, proteins)
>>> hits.is_sorted(by="key")
True


Use len to query the number of top hits, and the usual indexing notation to extract a particular Hit:

>>> len(hits)
1
>>> hits[0].name
b'938293.PRJEB85.HG003687_113'

__init__()

Create an empty TopHits instance.

compare_ranking(ranking)

Compare current top hits to previous top hits ranking.

This method is used by jackhmmer to record the hits obtained during each iteration, so that the inner loop can converge.

Parameters

ranking (KeyHash) – A keyhash containing the ranks of the top hits from a previous run.

Returns

int – The number of new hits found in this iteration.

New in version 0.6.0.

copy(pipeline)

Create a copy of this TopHits instance.

New in version 0.5.0.

is_sorted(by='key')

Check whether or not the hits are sorted with the given method.

See sort for a list of allowed values for the by argument.

merge(*others)

Concatenate the hits from this instance and others.

If the Z and domZ values used to compute E-values were computed by the Pipeline from the number of targets, the returned object will update them by summing self.Z and other.Z. If they were set manually, the manual value will be kept, provided both values are equal.

Returns

TopHits – A new collection of hits containing a copy of all the hits from self and other, sorted by key.

Raises

ValueError – When trying to merge together several hits obtained from different Pipeline with incompatible parameters.

Caution

This should only be done for hits obtained for the same domain on similarly configured pipelines. Some internal checks will be done to ensure this is not the case, but the results may not be consistent at all.

Example

>>> pli = Pipeline(thioesterase.alphabet)
>>> hits1 = pli.search_hmm(thioesterase, proteins[:1000])
>>> hits2 = pli.search_hmm(thioesterase, proteins[1000:2000])
>>> hits3 = pli.search_hmm(thioesterase, proteins[2000:])
>>> merged = hits1.merge(hits2, hits3)


New in version 0.5.0.

sort(by='key')

Sort hits in the current instance using the given method.

Parameters

by (str) – The comparison method to use to compare hits. Allowed values are: key (the default) to sort by key, or seqidx to sort by sequence index and alignment position.

to_msa(alphabet, *, sequences=None, traces=None, trim=False, digitize=False, all_consensus_cols=False)

Create multiple alignment of all included domains.

Parameters
• alphabet (Alphabet) – The alphabet of the HMM this TopHits was obtained from. It is required to convert back hits to single sequences.

• sequences (list of Sequence, optional) – A list of additional sequences to include in the alignment.

• traces (list of Trace, optional) – A list of additional traces to include in the alignment.

Keyword Arguments
• trim (bool) – Trim off any residues that get assigned to flanking $$N$$ and $$C$$ states (in profile traces) or $$I_0$$ and $$I_m$$ (in core traces).

• digitize (bool) – If set to True, returns a DigitalMSA instead of a TextMSA.

• all_consensus_cols (bool) – Force a column to be created for every consensus column in the model, even if it means having all gap character in a column.

Returns

MSA – A multiple sequence alignment containing the reported hits, either a TextMSA or a DigitalMSA depending on the value of the digitize argument.

New in version 0.3.0.

Changed in version 0.6.0: Added the sequences and traces arguments.

E

The per-target E-value threshold for reporting a hit.

New in version 0.5.0.

Type

float

T

The per-target score threshold for reporting a hit.

New in version 0.5.0.

Type
Z

The effective number of targets searched.

Type

float

bit_cutoffs

The model-specific thresholding option, if any.

New in version 0.5.0.

Type
block_length

The block length these hits were obtained with.

Is always None when the hits were not obtained from a long targets pipeline.

New in version 0.5.0.

Type
domE

The per-domain E-value threshold for reporting a hit.

New in version 0.5.0.

Type

float

domT

The per-domain score threshold for reporting a hit.

New in version 0.5.0.

Type
domZ

The effective number of significant targets searched.

Type

float

hits_included

The number of hits that are above the inclusion threshold.

Changed in version 0.5.0: Renamed from included to hits_included.

Type

int

hits_reported

The number of hits that are above the reporting threshold.

Changed in version 0.5.0: Renamed from reported to hits_reported.

Type

int

incE

The per-target E-value threshold for including a hit.

New in version 0.5.0.

Type

float

incT

The per-target score threshold for including a hit.

New in version 0.4.8.

Type
incdomE

The per-domain E-value threshold for including a hit.

New in version 0.5.0.

Type

float

incdomT

The per-domain score threshold for including a hit.

New in version 0.5.0.

Type
long_targets

Whether these hits were produced by a long targets pipeline.

New in version 0.5.0.

Type

bool

searched_models

The number of models searched.

New in version 0.5.0.

Type

int

searched_nodes

The number of model nodes searched.

New in version 0.5.0.

Type

int

searched_residues

The number of residues searched.

New in version 0.5.0.

Type

int

searched_sequences

The number of sequences searched.

New in version 0.5.0.

Type

int

strand

The strand these hits were obtained from.

Is always None when the hits were not obtained from a long targets pipeline, or when the long targets pipeline was configured to search both strands.

New in version 0.5.0.

Type

Hit¶

class pyhmmer.plan7.Hit

A high-scoring database hit found by the comparison pipeline.

is_dropped()

Check if the hit was dropped.

is_duplicate()

Check if the hit is a duplicate.

is_included()

Check if the hit should be included with respect to the thresholds.

is_new()

Check if the hit is a new hit.

is_reported()

Check if the hit should be reported with respect to the thresholds.

manually_drop()

Mark this hit as dropped.

Dropping a hit manually means that it will not be used when building a multiple sequence alignment from the TopHits object, even if it was above inclusion thresholds. This can be useful when manually selecting hits during an iterative search performed by Pipeline.iterate_seq.

New in version 0.6.0.

manually_include()

Mark this hit as included.

Including a hit manually means that it will be used when building a multiple sequence alignment from the TopHits object, even if it was under inclusion thresholds.

New in version 0.6.0.

accession

The accession of the database hit, if any.

Type
best_domain

The best scoring domain in this hit.

New in version 0.4.2.

Type

Domain

bias

The null2 contribution to the uncorrected score.

Type

float

description

The description of the database hit, if any.

Type
domains

The list of domains aligned to this hit.

Type

Domains

evalue

The e-value of the hit.

Type

float

name

The name of the database hit.

Type

bytes

pre_score

Bit score of the sequence before null2 correction.

Type

float

pvalue

The p-value of the bitscore.

New in version 0.4.2.

Type

float

score

Bit score of the sequence with all domains after correction.

Type

float

sum_score

Bit score reconstructed from the sum of domain envelopes.

New in version 0.4.6.

Type

float

Domains¶

class pyhmmer.plan7.Domains

A read-only view over the domains of a single Hit.

hit

The target hit these domains belong hit.

Type

Hit

Domain¶

class pyhmmer.plan7.Domain

A single domain in a query Hit.

hit

The hit this domains is part of.

Type

Hit

alignment

The alignment of this domain to a target sequence.

Type

Alignment

bias

The null2 score contribution to the domain score.

Type

float

c_evalue

The conditional e-value for the domain.

Type

float

correction

The null2 score when calculating a per-domain score.

Type

float

env_from

The start coordinate of the domain envelope.

Type

int

env_to

The end coordinate of the domain envelope.

Type

int

envelope_score

The forward score in the envelope, without null2 correction.

Type

float

i_evalue

The independent e-value for the domain.

Type

float

pvalue

The p-value of the domain bitscore.

Type

float

score

The overall score in bits, null2-corrected.

Type

float

Alignment¶

class pyhmmer.plan7.Alignment

An alignment of a sequence to a profile.

domain

The domain this alignment corresponds to.

Type

Domain

hmm_accession

The accession of the query, or its name if it has none.

New in version 0.1.4.

Type

bytes

hmm_from

The start coordinate of the alignment in the query HMM.

Type

int

hmm_name

The name of the query HMM.

Type

bytes

hmm_sequence

The sequence of the query HMM in the alignment.

Type

str

hmm_to

The end coordinate of the alignment in the query HMM.

Type

int

identity_sequence

The identity sequence between the query and the target.

Type

str

target_from

The start coordinate of the alignment in the target sequence.

Type

int

target_name

The name of the target sequence.

Type

bytes

target_sequence

The sequence of the target sequence in the alignment.

Type

str

target_to

The end coordinate of the alignment in the target sequence.

Type

int

Traces¶

TraceAligner¶

class pyhmmer.plan7.TraceAligner

A factory for aligning several sequences to a reference model.

Example

>>> aligner = TraceAligner()
>>> traces = aligner.compute_traces(thioesterase, proteins[:100])
>>> msa = aligner.align_traces(thioesterase, proteins[:100], traces)


New in version 0.4.7.

align_traces(hmm, sequences, traces, trim=False, digitize=False, all_consensus_cols=False)

Compute traces for a collection of sequences relative to an HMM.

Parameters
• hmm (HMM) – The reference HMM to use for the alignment.

• sequences (collection of DigitalSequence) – The sequences to align to the HMM.

• traces (Traces) – The traces corresponding to the alignment of sequences to hmm, obtained by a previous call to compute_traces.

• trim (bool) – Trim off any residues that get assigned to flanking $$N$$ and $$C$$ states (in profile traces) or $$I_0$$ and $$I_m$$ (in core traces).

• digitize (bool) – If set to True, returns a DigitalMSA instead of a TextMSA.

• all_consensus_cols (bool) – Force a column to be created for every consensus column in the model, even if it means having all gap character in a column. Note that this is enabled by default for hmmalign since HMMER v3.1, but disabled here.

Returns

MSA – A multiple sequence alignment containing the aligned sequences, either a TextMSA or a DigitalMSA depending on the value of the digitize argument.

Raises

AlphabetMismatch – when the alphabet of any of the sequences does not correspond to the HMM alphabet.

compute_traces(hmm, sequences)

Compute traces for a collection of sequences relative to an HMM.

Parameters
Returns

Traces – The traces for each sequence.

Raises

AlphabetMismatch – when the alphabet of any of the sequences does not correspond to the HMM alphabet.

Traces¶

class pyhmmer.plan7.Traces

A list of tracebacks obtained by aligning several sequences to a model.

New in version 0.4.7.

Trace¶

class pyhmmer.plan7.Trace

A traceback for the alignment of a model to a sequence.

New in version 0.4.7.

__init__(posteriors=False)

Create a new Trace object.

Parameters

posteriors (bool) – Whether or not to allocate additional memory for the storage of posterior probabilties.

expected_accuracy()

Returns the sum of the posterior residue decoding probabilities.

from_sequence(sequence)

Create a faux trace from a single sequence.

New in version 0.6.0.

L

The sequence length.

Type

int

M

The model length.

Type

int

posterior_probabilities

The posterior probability of each residue.

Type

VectorF or None

Iterative Searches¶

IterativeSearch¶

class pyhmmer.plan7.IterativeSearch

A helper class for running iterative queries like JackHMMER.

See Pipeline.iterate_seq and Pipeline.iterate_hmm for more information.

pipeline

The pipeline object to use to get hits on each iteration.

Type

Pipeline

builder

The builder object for converting multiple sequence alignments obtained after each run to a HMM.

Type

Builder

query

The query object to use for the first iteration.

Type
converged

Whether the iterative search already converged or not.

Type

bool

targets

The search targets to search for homologous sequences.

Type

PipelineSearchTargets

ranking

A mapping storing the rank of hits from previous iterations.

Type

KeyHash

iteration

The index of the last iteration done so far.

Type

int

Yields

IterationResult – A named tuple containing the hits, multiple sequence alignment and HMM for each iteration, as well as the iteration index and a flag marking whether the search converged.

References

• Johnson, Steven L., Eddy, Sean R. & Portugaly, Elon. Hidden Markov model speed heuristic and iterative HMM search procedure. BMC Bioinformatics 11, 431 (18 August 2010). doi:10.1186/1471-2105-11-431.

__init__(*args, **kwargs)

IterationResult¶

class pyhmmer.plan7.IterationResult

The results of a single iteration from an IterativeSearch.

hmm

The HMM used to search for hits during this iteration.

Type

HMM

hits

The hits found during this iteration.

Type

TopHits

msa

A multiple sequence alignment containing the hits from this iteration.

Type

DigitalMSA

converged

A flag marking whether this iteration converged (no new hit found in the target sequences with respect to the pipeline inclusion thresholds).

Type

bool

iteration

The number of iterations done so far. First iteration starts at 1.

Type

int

Miscellaneous¶

Cutoffs¶

class pyhmmer.plan7.Cutoffs

A mutable view over the score cutoffs of a HMM or a Profile.

New in version 0.4.6.

as_vector()

Return a view over the score cutoffs as a VectorF.

gathering_available()

Check whether the gathering thresholds are available.

noise_available()

Check whether the noise cutoffs are available.

trusted_available()

Check whether the trusted cutoffs are available.

gathering

The gathering thresholds, if any.

Example

This property can be used to set the gathering cutoffs by passing it an iterable of two float:

>>> thioesterase.cutoffs.gathering = (180.0, 120.0)
>>> thioesterase.cutoffs.gathering_available()
True
>>> thioesterase.cutoffs.gathering
(180.0, 120.0)


Set the attribute to None or delete it with del to clear the gathering thresholds:

>>> thioesterase.cutoffs.gathering = None
>>> thioesterase.cutoffs.gathering_available()
False


New in version 0.4.8.

Type
gathering1

The first gathering threshold, if any.

Type
gathering2

The second gathering threshold, if any.

Type
noise

The noise cutoffs, if available.

New in version 0.4.8.

Type
noise1

The first noise cutoff, if any.

Type
noise2

The second noise cutoff, if any.

Type
trusted

The trusted cutoffs, if available.

New in version 0.4.8.

Type
trusted1

The first trusted score cutoff, if any.

Type
trusted2

The second trusted score cutoff, if any.

Type

EvalueParameters¶

class pyhmmer.plan7.EvalueParameters

A mutable view over the e-value parameters of a HMM or a Profile.

The e-value for each filter is estimated based off a maximum likelihood distribution fitted for each profile HMM, either a Gumbel distribution for the MSV and Viterbi filters, or an exponential distribution for the Forward filter.

New in version 0.4.6.

as_vector()

Return a view over the e-value parameters as a VectorF.

f_lambda

The $$\lambda$$ parameter for the Forward filter distribution.

Type
f_tau

The $$\tau$$ parameter for the Forward filter distribution.

Type
m_lambda

The $$\lambda$$ parameter for the MSV filter distribution.

Type
m_mu

The $$\mu$$ parameter for the MSV filter distribution.

Type
v_lambda

The $$\lambda$$ parameter for the Viterbi filter distribution.

Type
v_mu

The $$\mu$$ parameter for the Viterbi filter distribution.

Type

Offsets¶

class pyhmmer.plan7.Offsets

A mutable view over the disk offsets of a profile.

PipelineSearchTargets¶

class pyhmmer.plan7.PipelineSearchTargets

An optimized storage of search target sequences for a Pipeline.

To pass the target sequences efficiently in Pipeline.search_hmm, an array is allocated so that the inner loop can iterate over the target sequences without having to acquire the GIL for each new sequence (this gave a huge performance boost in v0.4.5). However, there was no way to reuse this between different queries; some memory recycling was done, but the target sequences had to be indexed for every query.

This class allows the reference array to be maintained between several queries. Internally, Pipeline.search_hmm will build one from any other argument type; but if a PipelineSearchTargets is passed it will be used as-is in the search loop.

alphabet

The biological alphabet shared by all sequences in the search targets.

Type

Alphabet, readonly

New in version 0.5.0.

__init__(sequences)

Create a new list of search targets.

Parameters

sequence (iterable of DigitalSequence) – An iterable of sequences stored in digital mode to use as targets for a search pipeline.

Raises

AlphabetMismatch – When all sequences don’t have the same Alphabet.