SCORCH2: A Generalized Heterogeneous Consensus Model for High-Enrichment Interaction-Based Virtual Screening.

Area of Science:

• Computational chemistry
• Machine learning in drug discovery
• Bioinformatics

Background:

• Drug discovery is complex, costly, and time-consuming, with high failure rates.
• Identifying hit compounds with moderate affinity for biological targets is a key bottleneck.
• Current in silico virtual screening methods face limitations like overfitting, data bias, and poor interpretability.

Purpose of the Study:

• To introduce SCORCH2, a machine learning framework for enhanced virtual screening.
• To improve both the performance and interpretability of virtual screening processes.
• To address limitations of existing in silico screening methods.

Main Methods:

• Developed SCORCH2, a machine learning-based framework leveraging interaction features.
• Evaluated SCORCH2's performance against its predecessor, SCORCH.
• Assessed SCORCH2's hit identification capabilities on novel biological targets.

Main Results:

• SCORCH2 demonstrates superior predictive accuracy and generalizability across diverse biological targets compared to SCORCH.
• SCORCH2 shows robust hit identification on previously unseen targets, indicating strong transferability.
• The framework streamlines the screening process by removing the need for docking pose selection.

Conclusions:

• SCORCH2 offers enhanced performance and interpretability in virtual screening.
• The framework shows significant potential for accelerating early-stage drug discovery.
• SCORCH2 represents a valuable advancement in computational drug discovery tools.