DrugForm-TAS: Target-Agnostic Selectivity as Proteome-wide Binding Propensity Estimation

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Assessing small molecule-protein binding selectivity is crucial but computationally challenging.
• Current methods require extensive affinity predictions across numerous targets, limiting large-scale screening.

Purpose of the Study:

• Introduce DrugForm-TAS, the first model for direct, quantitative prediction of proteome-wide binding propensity.
• Enable target-agnostic selectivity profiling for efficient early-stage drug design.

Main Methods:

• Utilized a lightweight transformer-like neural network trained on BindingDB data.
• Operates solely on ligand SMILES representations, requiring no prior target knowledge.
• Built upon the DrugForm-DTA model for affinity prediction.

Main Results:

• DrugForm-TAS provides a direct, unconditional measure of binding propensity.
• Demonstrated correlation between model predictions and experimental observations.
• Significantly reduces candidate sets by acting as a fast pre-screening filter.

Conclusions:

• DrugForm-TAS enables efficient, target-agnostic nonspecificity profiling.
• Offers a novel tool for rapid selectivity estimation in early drug design.
• Accelerates virtual screening and de novo molecular design processes.