Methods and systems for determining gene sets for diagnosis and treatment of disease states

A method for obtaining a gene set capable of classifying a patient between a first state and a second state, the method comprising:

• a) training a first machine learning model based on a first data set comprising or derived from gene expression measurements of genes within a plurality of gene modules, to classify a plurality of reference patients between the first state and the second state, wherein the gene expression measurements are obtained from a plurality of reference biological samples derived or obtained from the plurality of reference patients;
• b) selecting N gene modules from the plurality of gene modules based on feature importance values and/or absolute SHAP values of input features of the first machine learning model for classifying the plurality of reference patients between the first state and the second state, wherein Nis an integer;
• c) optionally removing redundant genes from genes within the N gene modules to obtain a first gene set;
• d) training a second machine learning model based on a second data set comprising gene expression measurements of genes within the first gene set to classify the plurality of reference patients between the first state and the second state;
• e) selecting M genes from the genes within the first gene set based on feature importance values and/or absolute SHAP values of input features of the second machine learning model for classifying the plurality of reference patients between the first state and the second state, wherein M is an integer
• wherein the selected M genes form the gene set capable of classifying the patient between the first state and the second state, and wherein a first portion of the plurality of reference patients belong to the first state and a second portion of the plurality of reference patients belong to the second state.