-
Notifications
You must be signed in to change notification settings - Fork 15
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Add IM2Deep feature generator (#121)
* im2deep * models + debug * fix * add models * fixes in im2deep.py * fixes in im2deep.py * im2deep implementation * CCS shift calculation fix * models + plot * add unused argument * IM2Deep plot correct labels * calibrate per charge option * new models and reference * correct format reference * fix model name in IM2Deep * change use of reference dataset * IM2Deep calibrate correctly * Implement IM2Deep package * GUI changes im2deep * add im2deep to pyproject * change dockerfile * optimise spectrum parsing * update toml file * added debug logging * requested changes * change numpy version upper limit * fix bug * Use newer setup-python action; revert numpy upper limit * Fix numpy version for py311 * Update pyproject.toml Use semi-colon instead of comma's in dependencies * requested changes * Update ms2rescore/feature_generators/im2deep.py --------- Co-authored-by: rodvrees <[email protected]> Co-authored-by: rodvrees <[email protected]> Co-authored-by: rodvrees <[email protected]> Co-authored-by: RalfG <[email protected]>
- Loading branch information
5 people
authored
Feb 21, 2024
1 parent
d8340f5
commit 667822e
Showing
12 changed files
with
416 additions
and
106 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,228 @@ | ||
| """ | ||
| IM2Deep ion mobility-based feature generator. | ||
| IM2Deep is a fully modification-aware peptide ion mobility predictor. It uses a deep convolutional | ||
| neural network to predict retention times based on the atomic composition of the (modified) amino | ||
| acid residues in the peptide. See | ||
| `github.com/compomics/IM2Deep <https://github.com/compomics/IM2Deep>`_ for more information. | ||
| """ | ||
|
|
||
| import contextlib | ||
| import logging | ||
| import os | ||
| from inspect import getfullargspec | ||
| from itertools import chain | ||
| from typing import List, Optional | ||
|
|
||
| import numpy as np | ||
| from im2deep.im2deep import predict_ccs | ||
| from psm_utils import PSMList | ||
|
|
||
| from ms2rescore.feature_generators.base import FeatureGeneratorBase | ||
|
|
||
| os.environ["TF_CPP_MIN_LOG_LEVEL"] = "2" | ||
| logger = logging.getLogger(__name__) | ||
|
|
||
|
|
||
| class IM2DeepFeatureGenerator(FeatureGeneratorBase): | ||
| """IM2Deep collision cross section feature generator.""" | ||
|
|
||
| def __init__( | ||
| self, | ||
| *args, | ||
| lower_score_is_better: bool = False, | ||
| spectrum_path: Optional[str] = None, | ||
| processes: int = 1, | ||
| calibrate_per_charge: bool = True, | ||
| **kwargs, | ||
| ): | ||
| """ | ||
| Initialize the IM2DeepFeatureGenerator. | ||
| Parameters | ||
| ---------- | ||
| lower_score_is_better : bool, optional | ||
| A boolean indicating whether lower scores are better for the generated features. | ||
| spectrum_path : str or None, optional | ||
| Optional path to the spectrum file used for IM2Deep predictions. | ||
| processes : int, optional | ||
| Number of parallel processes to use for IM2Deep predictions. | ||
| calibrate_per_charge : bool, optional | ||
| A boolean indicating whether to calibrate CCS values per charge state. | ||
| **kwargs : dict, optional | ||
| Additional keyword arguments. | ||
| Returns | ||
| ------- | ||
| None | ||
| """ | ||
| super().__init__(*args, **kwargs) | ||
| self.lower_score_is_better = lower_score_is_better | ||
| self.spectrum_path = spectrum_path | ||
| self.processes = processes | ||
| self.deeplc_kwargs = kwargs or {} | ||
|
|
||
| self._verbose = logger.getEffectiveLevel() <= logging.DEBUG | ||
|
|
||
| # Lazy-load DeepLC | ||
| from deeplc import DeepLC | ||
|
|
||
| self.im2deep = DeepLC | ||
|
|
||
| # Remove any kwargs that are not DeepLC arguments | ||
| self.im2deep_kwargs = { | ||
| k: v for k, v in self.deeplc_kwargs.items() if k in getfullargspec(DeepLC).args | ||
| } | ||
| self.im2deep_kwargs.update({"config_file": None}) | ||
|
|
||
| # TODO: Implement im2deep_retrain? | ||
|
|
||
| self.im2deep_predictor = None | ||
| self.calibrate_per_charge = calibrate_per_charge | ||
|
|
||
| @property | ||
| def feature_names(self) -> List[str]: | ||
| return [ | ||
| "ccs_observed_im2deep", | ||
| "ccs_predicted_im2deep", | ||
| "ccs_error_im2deep", | ||
| "abs_ccs_error_im2deep", | ||
| "perc_ccs_error_im2deep", | ||
| ] | ||
|
|
||
| def add_features(self, psm_list: PSMList) -> None: | ||
| """Add IM2Deep-derived features to PSMs""" | ||
|
|
||
| logger.info("Adding IM2Deep-derived features to PSMs") | ||
|
|
||
| # Get easy-access nested version of PSMlist | ||
| psm_dict = psm_list.get_psm_dict() | ||
|
|
||
| # Run IM2Deep for each spectrum file | ||
| current_run = 1 | ||
| total_runs = sum(len(runs) for runs in psm_dict.values()) | ||
|
|
||
| for runs in psm_dict.values(): | ||
| # Reset IM2Deep predictor for each collection of runs | ||
| self.im2deep_predictor = None | ||
| self.selected_model = None | ||
| for run, psms in runs.items(): | ||
| logger.info( | ||
| f"Running IM2Deep for PSMs from run ({current_run}/{total_runs}): `{run}`..." | ||
| ) | ||
|
|
||
| # Disable wild logging to stdout by TensorFlow, unless in debug mode | ||
| with ( | ||
| contextlib.redirect_stdout(open(os.devnull, "w")) | ||
| if not self._verbose | ||
| else contextlib.nullcontext() | ||
| ): | ||
| # Make new PSM list for this run (chain PSMs per spectrum to flat list) | ||
| psm_list_run = PSMList(psm_list=list(chain.from_iterable(psms.values()))) | ||
|
|
||
| logger.debug("Calibrating IM2Deep...") | ||
|
|
||
| # Convert ion mobility to CCS and calibrate CCS values | ||
| psm_list_run_df = psm_list_run.to_dataframe() | ||
| psm_list_run_df["charge"] = [ | ||
| peptidoform.precursor_charge | ||
| for peptidoform in psm_list_run_df["peptidoform"] | ||
| ] | ||
| psm_list_run_df["ccs_observed"] = psm_list_run_df.apply( | ||
| lambda x: self.im2ccs( | ||
| x["ion_mobility"], | ||
| x["precursor_mz"], # TODO: Why does ionmob use calculated mz? | ||
| x["charge"], | ||
| ), | ||
| axis=1, | ||
| ) | ||
|
|
||
| # Create dataframe with high confidence hits for calibration | ||
| cal_psm_df = self.make_cal_df(psm_list_run_df) | ||
|
|
||
| # Make predictions with IM2Deep | ||
| logger.debug("Predicting CCS values...") | ||
| calibrated_predictions = predict_ccs( | ||
| psm_list_run, cal_psm_df, write_output=False | ||
| ) | ||
|
|
||
| # Add features to PSMs | ||
| logger.debug("Adding features to PSMs...") | ||
| predictions = calibrated_predictions | ||
| observations = psm_list_run_df["ccs_observed"] | ||
| ccs_diffs_run = np.abs(predictions - observations) | ||
| for i, psm in enumerate(psm_list_run): | ||
| psm["rescoring_features"].update( | ||
| { | ||
| "ccs_observed_im2deep": observations[i], | ||
| "ccs_predicted_im2deep": predictions[i], | ||
| "ccs_error_im2deep": ccs_diffs_run[i], | ||
| "abs_ccs_error_im2deep": np.abs(ccs_diffs_run[i]), | ||
| "perc_ccs_error_im2deep": np.abs(ccs_diffs_run[i]) | ||
| / observations[i] | ||
| * 100, | ||
| } | ||
| ) | ||
|
|
||
| current_run += 1 | ||
|
|
||
| def im2ccs(self, reverse_im, mz, charge, mass_gas=28.013, temp=31.85, t_diff=273.15): | ||
| """ | ||
| Convert ion mobility to CCS. | ||
| Parameters | ||
| ---------- | ||
| reverse_im : float | ||
| Reduced ion mobility. | ||
| mz : float | ||
| Precursor m/z. | ||
| charge : int | ||
| Precursor charge. | ||
| mass_gas : float, optional | ||
| Mass of gas, by default 28.013 | ||
| temp : float, optional | ||
| Temperature in Celsius, by default 31.85 | ||
| t_diff : float, optional | ||
| Factor to convert Celsius to Kelvin, by default 273.15 | ||
| Notes | ||
| ----- | ||
| Adapted from theGreatHerrLebert/ionmob (https://doi.org/10.1093/bioinformatics/btad486) | ||
| """ | ||
|
|
||
| SUMMARY_CONSTANT = 18509.8632163405 | ||
| reduced_mass = (mz * charge * mass_gas) / (mz * charge + mass_gas) | ||
| return (SUMMARY_CONSTANT * charge) / ( | ||
| np.sqrt(reduced_mass * (temp + t_diff)) * 1 / reverse_im | ||
| ) | ||
|
|
||
| # TODO: replace threshold by identified psms? | ||
| def make_cal_df(self, psm_list_df, threshold=0.95): | ||
| """Make dataframe for calibration of IM2Deep predictions. | ||
| Parameters | ||
| ---------- | ||
| psm_list_df : pd.DataFrame | ||
| DataFrame with PSMs. | ||
| threshold : float, optional | ||
| Threshold for high confidence hits, by default 0.95. | ||
| Returns | ||
| ------- | ||
| pd.DataFrame | ||
| DataFrame with high confidence hits for calibration.""" | ||
|
|
||
| psm_list_df = psm_list_df[ | ||
| psm_list_df["charge"] < 5 | ||
| ] # predictions do not go higher for IM2Deep | ||
| high_conf_hits = list( | ||
| psm_list_df["spectrum_id"][psm_list_df["score"].rank(pct=True) > threshold] | ||
| ) | ||
| logger.debug( | ||
| f"Number of high confidence hits for calculating shift: {len(high_conf_hits)}" | ||
| ) | ||
| # Filter df for high_conf_hits | ||
| cal_psm_df = psm_list_df[psm_list_df["spectrum_id"].isin(high_conf_hits)] | ||
| return cal_psm_df |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.