TumFlow: An AI Model for Predicting New Anticancer Molecules.

Area of Science:

• Oncology
• Computational Chemistry
• Artificial Intelligence in Drug Discovery

Background:

• Melanoma is a prevalent cancer, and traditional drug discovery is slow and expensive.
• Artificial Intelligence (AI) offers a solution to expedite the identification and evaluation of potential drug candidates.
• Machine learning, specifically normalizing flow models, shows promise for generating novel therapeutic molecules.

Purpose of the Study:

• To introduce TumFlow, a novel AI model for generating new molecular entities for cancer therapy.
• To leverage AI for accelerating the discovery of effective anticancer drugs.
• To address limitations of conventional generative models by ensuring synthetic feasibility.

Main Methods:

• Developed and trained TumFlow, an AI model utilizing normalizing flows.
• Utilized the NCI-60 dataset, focusing on the melanoma SK-MEL-28 cell line.
• Applied TumFlow to generate novel molecules and optimize existing melanoma drug candidates.

Main Results:

• TumFlow successfully generated novel molecules with predicted enhanced efficacy against tumor growth.
• The generated molecules were designed to be synthetically feasible, overcoming a key challenge in drug discovery.
• Optimization of known melanoma drugs yielded new compounds with predicted improved effectiveness.

Conclusions:

• TumFlow represents a significant advancement in AI-driven drug discovery for cancer.
• The model's ability to generate synthetically feasible and potentially more effective molecules accelerates therapeutic development.
• This AI approach holds promise for discovering novel, undocumented melanoma treatments.