SimPathFinder

Project objective

• 1. Predicting enzymatic pathway type/class from partially/fully annotated enzymes**.
• 2. Identify similar, existing pathways from KEGG and MetaCyc, given a set of partially/fully annotated enzymes**.

**from a prokaryotic/eukaryotic organism

List of files and directories:

• ecdata contains all relevant unlabeled data from BioCyc and MetaCyc (scrapped from webpages && downloaded from API)

• labeldata contains all relevant labeled data from KEGG and MetaCyc

• src contains all the modeling/scrapping scripts

• Flaskapp contains flask templates and scripts for webserver

• tests contains unit tests

List of classes:

Classes for data extraction and preparation

• ExtractUnlabeledData class to extract unlabeled data from BioCyc and MetaCyc

• SampleUnlabeledData class to expand unlabeled data using resampling, child of ExtractUnlabeledData

• PathwayScrapper class to extract labeled data from MetaCyc

• ExtractLabeledData class to process and clean labeled data from MetaCyc webpages

• BalanceLabelData class to balance labeled data

Classes to create embeddings

• CreateEmbedding class to create fasttext embeddings

• ParameterizeEmbedding class to explore hyperparameters for creating embeddigns, child of CreateEmbedding

• ClusterEmbedding class to cluster enzyme embedding vectors, child of CreateEmbedding

• ClusterPWYEmbedding class to cluster pathway embedding vectors, child of CreateEmbedding

Classes to create ML models

• ModelVectors class to get dependent and independent variables for building models

• BuildClassicalModel class to set up and run classical ML models using scikit-learn, child of ModelVectors

• BuildControlModels class to set up and run classical ML models on control datasets, child of BuildClassicalModel

• BuildAnnot3Models class to set up and run classical ML models on datasets build using 3 digit EC number annotations, child of BuildClassicalModel

• BuildAnnot2Models class to set up and run classical ML models on datasets build using 2 digit EC number annotations, child of BuildClassicalModel

• BuildAnnot1Models class to set up and run classical ML models on datasets build using 1 digit EC number annotations, child of BuildClassicalModel

Classes to create DNN models

• DNNModel class to create, parameterize and test LSTM based neural network model, child of ModelVectors

Classes to analyze/measure results

• Metrics class to define all required metrics

• EvaluateMetrics class evaluate output from ML model, child of Metrics and ModelVectors

• EvaluateControl class evaluate output from ML model for control datasets, child of EvaluateMetrics

• EvaluateAnnot3 class to evaluate output from ML model for datasets build using 3 digit EC number annotations, child of EvaluateMetrics

• EvaluateAnnot2 class to evaluate output from ML model for datasets build using 2 digit EC number annotations, child of EvaluateMetrics

• EvaluateAnnot1 class to evaluate output from ML model for datasets build using 1 digit EC number annotations, child of EvaluateMetrics

• combinedEvaluations class to compare and contrast output from all datasets, child of Metrics and ModelVectors

• MetricsDNN class to define all required metrics for NN output

• EvaluateMetricsDNN class evaluate output from DNN model, child of MetricsDNN and ModelVectors

• EvaluateMetricsAnnot3DNN class evaluate output from DNN model on datasets build using 1 digit EC number annotations, child of EvaluateMetricsDNN

Classes to explore model performance on KEGG dataset (validation)

• ExtractKEGGData class to extract and evaluate KEGG data

• ExtractKEGGDataControl class to evalate KEGG data on control models, child of ExtractKEGGData

Web server

• Web server

Information

Paper: Predicting ontology of enzymatic pathways using transfer learning

Author(s): Sai J. Ganesan, Andrej Sali

Citation: TBD