Use of AI models for compound design- QC-351Discipline(s) souhaitée: Génie - informatique / électrique, Génie, Informatique, Sciences mathématiques, Mathématiques, Statistiques / études actuarielles
Entreprise: Servier Canada
Durée du projet: 6 months to 1 year
Preferred start date: As soon as possible.
Langue exigée: Flexible
Emplacement(s): Laval, Montreal, QC, Canada; Canada
Nombre de postes: 2
Au sujet de l’entreprise:
Servier Canada Inc is the Canadian affiliate of The Servier Research Group, a leading French research based organisation, specializing in ethical pharmaceuticals.
Servier is an organisation as well as an international foundation involved in research activities which are focused on further improving health and bringing better care to patients.
Medicines add years to life and life to years as well as offer patients a better quality of life. In most cases, the cost of treating a disease is lower than the cost of not treating and the benefits provided by the treatments are often priceless to patients. This applies to the majority of common chronic diseases such as hypertension or diabetes.
These two diseases can, in fact, lead to severe complications which treatment with drugs can now prevent. This proves that although medicines may represent an immediate cost for the patient and the community, they remain an investment in duration and quality of life.
Servier’s Laboratories are transforming their way to research and develop drugs. The goal is to reimagine medicine using Data driven and Artificial intelligence (AI) approaches. To succeed, a new AI Hub in Montreal is created.
Veuillez décrire le projet.:
Absorption, Distribution, Metabolism and Excretion and Toxicity (ADMET) are key parameters that need to be considered in the design of all new drugs and form part of a multi parameter optimization paradigm that faces the Medicinal Chemist in the design cycle.
The aim of this part will be to propose a new and efficient way to predict ADME Tox (Absorption, Distribution, Metabolism and Excretion and Toxicity) properties, key parameters needed through the design cycle of a drug. Today, ADME Tox in silico models are created using conventional machine learning (for example: random forest based) and classical molecular descriptors. This approach is limited to the information linked to only one property (e.g. activity) at a time and chemical description may not be optimun for automated drug proposals.
The goal is to form part of a fully automated in silico toolkit for multi objective lead optimisation that can provide additional resources to augment the internal AI efforts and will be tested through synthesis and ADMET tests on designed compounds.
Part2: Molecule generator
Developing de novo algorithms to design and predict active compounds with desired properties has the potential to reduce the time and the cost of finding high quality compounds for further progression.
Current existing machine learning (AI) technologies for compound design rely onwell established technologies, namely recurrent neural networks and variational autoencoder, based on the SMILES and activity model of the molecule. Whilst this can be an effective approach, they have significant limitations, they often produce synthetically intractable molecules or even invalid chemistry output.
The aim of this part will be to propose new molecules with optimized properties to the medicinal chemist in Servier, so they could apply the de novo molecule generation to obtain new molecular scaffolds asideas for a back up project.
Expertise ou compétences exigées:
Deep Learning, Machine Learning, Data Science, Bioinformatics, Meta-learing, Multi-task learning, Embeddings, Molecules features, Generative models, GANs, AUTO-ENCODER, Python