System and method for processing experimental data
The method for processing experimental data can include: determining experimental data (e.g., mass spectrometry spectra) and processing the experimental data. In variants, processing the experimental data can include: identifying one or more molecules, comparing experimental samples, determining a quantification, evaluating a quality of the experimental data, and/or otherwise processing the experimental data. The method can optionally include determining supplemental information, determining a set of candidate molecules, training a model, and/any other suitable steps.
1. A method for molecule identification, comprising:
training a first encoder using a known match between a training sample and a set of training spectra, the training sample comprising a set of training molecules, wherein training the first encoder comprises:
using the first encoder, determining a set of molecule predictions based on the set of training spectra, wherein a number of true-positive molecule predictions and a number of true-negative molecule predictions are determined based on a comparison between the set of molecule predictions and the set of training molecules;
determining a loss for the set of molecule predictions, the loss comprising an accuracy metric determined based on the number of true-positive molecule predictions and the number of true-negative molecule predictions; and
training the first encoder based on the loss;
determining a mass spectrometry spectrum for a molecule;
determining an embedding for the mass spectrometry spectrum, using the first encoder;
for each candidate molecule in a set of candidate molecules:
determining an embedding for the candidate molecule based on a sequence for the candidate molecule, using a second encoder; and
using a scoring model, determining a score for the candidate molecule based on the embedding for the mass spectrometry spectrum and the embedding for the candidate molecule; and
selecting a candidate molecule from the set of candidate molecules based on the scores.
2. The method of claim 1 , wherein determining the score for the candidate molecule comprises: determining a similarity metric based on the embedding for the mass spectrometry spectrum and the embedding for the candidate molecule; and determining the score based on the similarity metric.
3. The method of claim 2 , wherein the similarity metric comprises a distance between the embedding for the mass spectrometry spectrum and the embedding for the candidate molecule.
4. The method of claim 1 , wherein the first encoder comprises a CNN.
5. The method of claim 4 , wherein the first encoder further comprises a transformer.
6. The method of claim 1 , wherein the second encoder comprises a combination of an RNN and a transformer.
7. The method of claim 1 , wherein the mass spectrometry spectrum comprises an experimental dataset containing intensities across at least two dimensions, wherein a first dimension of the experimental dataset is mass to charge ratio (m/z) and a second dimension of the experimental dataset is retention time.
8. The method of claim 7 , wherein the experimental dataset contains intensities across a third dimension, wherein the third dimension is ion mobility.
9. The method of claim 1 , wherein the first encoder is configured to accommodate receiving as input experimental datasets with a variable number of dimensions.
10. The method of claim 1 , further comprising determining a location of a post-translational modification within the candidate molecule based on the embedding for the mass spectrometry spectrum and the embedding for the candidate molecule.
11. The method of claim 1 , wherein individual matches between each training spectrum in the set of training spectra and a corresponding training molecule in the set of training molecules is not known, wherein the set of molecule predictions is determined using the first encoder, the second encoder, and the scoring model.
12. The method of claim 11 , wherein the loss for the set of molecule predictions is further determined based on a second set of molecule predictions determined using at least two first-generation search engines.
13. The method of claim 1 , wherein the first encoder, the second encoder, and the scoring model are not trained using decoy molecules.
14. The method of claim 1 , wherein the embedding for the mass spectrometry spectrum is determined based on the mass spectrometry spectrum and supplemental experimental information, wherein the supplemental experimental information comprises at least one of: instrument type or fragmentation method.
15. A method for molecule identification, comprising:
training an encoder using a known match between a training sample and a set of training spectra, the training sample comprising a set of training molecules, wherein training the encoder comprises:
using the encoder, determining a set of molecule predictions based on the set of training spectra, wherein a number of true-positive molecule predictions and a number of true-negative molecule predictions are determined based on a comparison between the set of molecule predictions and the set of training molecules;
determining a loss for the set of molecule predictions, the loss comprising an accuracy metric determined based on the number of true-positive molecule predictions and the number of true-negative molecule predictions; and
training the encoder based on the loss;
determining a first set of mass spectrometry spectra for a first set of molecules in a first sample;
determining a second set of mass spectrometry spectra for a second set of molecules in a second sample;
determining an embedding for each mass spectrometry spectrum in the first set of mass spectrometry spectra, using the encoder;
determining an embedding for each mass spectrometry spectrum in the second set of mass spectrometry spectra, using the encoder;
clustering the embeddings for each mass spectrometry spectrum in the first and second sets of mass spectrometry spectra into a set of clusters;
identifying an embedding of interest based on the set of clusters; and
identifying a molecule based on the embedding of interest, wherein the molecule is present in the second sample and is not present in the first sample.
16. The method of claim 15 , wherein the encoder encodes each mass spectrometry spectrum into an embedding space, wherein the embeddings are clustered within the embedding space.
17. The method of claim 15 , wherein each cluster in the set of clusters has a cluster purity of at least 50%.
18. The method of claim 15 , wherein identifying the molecule comprises:
for each candidate molecule in a set of candidate molecules:
determining an embedding for the candidate molecule based on a sequence for the candidate molecule, using a second encoder; and
using a scoring model, determining a score for the candidate molecule based on the embedding of interest and the embedding for the candidate molecule; and
selecting a candidate molecule from the set of candidate molecules based on the scores.
19. The method of claim 15 , wherein the first sample and the second sample comprise samples from a multiple-arm experiment.