DeepCarc

DeepCarc predicts the carcinogenic potential of small molecules using deep learning-based model-level representations.

Key Features:

• Deep Learning Integration: Leverages advanced deep learning techniques to analyze molecular data for carcinogenicity prediction.
• Model-Level Representation: Utilizes model-level representations to capture complex patterns in molecular data and improve predictive accuracy.
• Performance Metrics: Achieved a Matthews correlation coefficient (MCC) of 0.432 on a test set from the National Center for Toxicological Research liver cancer database (NCTRlcdb) with an average improvement of 37% over four advanced deep learning-powered QSAR models.

Scientific Applications:

• Carcinogenicity Assessment: Computational assessment of carcinogenic potential for small molecules.
• Drug Development and Environmental Chemistry: Screening compounds from databases such as DrugBank and Tox21 to identify potential carcinogens for drug development and environmental monitoring.

Methodology:

Developed using a dataset of 692 compounds and validated on a test set of 171 compounds; evaluated on the NCTRlcdb test set using Matthews correlation coefficient (MCC).