Novel dimensionality reduction approaches for vector-based representations of expression profiles and chemical compounds to assist in the development of acute myeloid leukemia therapies

Principal Investigator: Sébastien Lemieux
Theme : Artificial Intelligence (AI) / Health
Competition : Omics Data Against Cancer (ODAC) Competition
Status : Completed
Start : Oct. 1, 2020
End: Sept. 30, 2022
Budget : $300,000.00


New drugs have been introduced recently for treating Acute Myeloid Leukemia (AML) but their optimal use is still being refined with yet limited impact on long-term survival of older patients. Precision medicine, assisted by artificial intelligence (AI) algorithms applied to genomics-based tests, promises to optimally associate treatments to each patient. Significant efforts have been put in developing AI algorithms trained on transcriptomic profiles or on the chemical structure of drugs. Unfortunately, the level of accuracy of these algorithms is still insufficient to have a clinical impact. Here, we propose to shift attention from AI algorithms toward the development of novel representations for both expression profiles and chemical structure. Essentially, these algorithms require “questions” to be asked in the form of large numeric vectors. Transforming a catalogue of expressed genes, including personal genetic variations and cancer-specific mutations into a numeric vector require massive simplifications.


By introducing advanced techniques to prepare a more detailed representation of the disease, we expect a significant increase in the accuracy of AI algorithms. We plan to perform a similar exploration regarding the presentation of AI algorithms of chemical structures of AML drugs. The proposed project thus aims at increasing the accuracy of predicting the efficacy of a chemical compound to specifically act against a given AML based on its expression profile. Such algorithm would be invaluable to assist in the design of new classes of drugs for resistant subtypes of AML or to assist in the patient-specific selection of the most appropriate therapeutic agent.


Lead Genome Centre: Génome Québec

Partners: IVADO, Oncopole


Guy Sauvageau Université de Montréal/IRIC
Yoshua Bengio Université de Montréal/Mila