Using predictive machine learning models for drug response simulation by calibrating patient-specific pathway signatures

Abstract

The utility of pathway signatures lies in their capability to determine whether a specific pathway or biological process is dysregulated in a given patient. These signatures have been widely used in machine learning (ML) methods for a variety of applications including precision medicine, drug repurposing, and drug discovery. In this work, we leverage highly predictive ML models for drug response simulation in individual patients by calibrating the pathway activity scores of disease samples. Using these ML models and an intuitive scoring algorithm to modify the signatures of patients, we evaluate whether a given sample that was formerly classified as diseased, could be predicted as normal following drug treatment simulation. We then use this technique as a proxy for the identification of potential drug candidates. Furthermore, we demonstrate the ability of our methodology to successfully identify approved and clinically investigated drugs for four different cancers, outperforming six comparable state-of-the-art methods. We also show how this approach can deconvolute a drugs’ mechanism of action and propose combination therapies. Taken together, our methodology could be promising to support clinical decision-making in personalized medicine by simulating a drugs’ effect on a given patient.

Multi-task analysis of gene expression data on cancer public datasets

“…Their parameters for node and edge attributes in the first strategy as well as the balancing option of the second one was tested in two scenarios, using micro array and RNASeq data sources. The DReCaS workflow extended the methodology proposed in [14] by allowing the scoring matrix optimization and transfer learning from one experiment to another. We also standardized the procedure in a reproducible manner dealing since the raw expression values processing till the individual drug screening or custom drug combination response simulation.…”

“…We proposed a framework (ScreenDOP – Screening for Disease Outcome Prediction) that offers two strategies to extract knowledge from the expression data provided by the GEO database from NCBI that have the outcome information for all the samples and were designed for the disease outcome prediction task. In the second task, we developed a workflow (DReCaS – Drug Response Calibration Simulation) that applies gene expression data with gene set enrichment to predict drug response of samples to drugs automatizing the strategy proposed in [14] illustrating its application using liver cancer datasets from ICGC. Both workflows for the two tasks generate a report containing the log of the tasks, the date and time they were executed, the duration of the execution and the memory usage, to monitor the bottlenecks according to the datasets’ size.…”

“…We developed a workflow (DReCaS – Drug Response Calibration Simulation) to automatize the methodology for drug response calibration simulation proposed by [14], adding new functionalities to complement the original method and providing then an end-to-end workflow to perform drug response experiment screening. The workflow comprehends the following procedures: (i) Raw data processing, (ii) Model training, (iii) Weights optimization for the calibrated scoring matrix, (iv) evaluation and prioritization of individual drugs, and (v) drug combination evaluation.…”

“…This procedure is responsible for training the model that differs from healthy and disease samples, and calculates the modified pathway scores from the previous calculated table exactly as specified [14]. However, in the training phase we added a performance log to track all the most used performance metrics values (accuracy, precision, recall and f1), allowing model transparency at every step of the pipeline.…”

“…As we showed in the table the bad outcome yields expressive increase of genes correlated to each other showing potential dependencies from each other to guide personalized medicine [47]. Since the core method for drug response simulation was already extensively tested in [14], we evaluated the new functionalities proposed in this paper. Firstly, we made two experiments in the LIRI-JP dataset to test (1) all the types of normalization and…”