Curiosity-Driven AI for Drug Discovery: Integrating Intrinsic Motivation and Reflective Reasoning for Safe Molecular Innovation
Year: 2025
Type: article
Abstract: The escalating cost and lengthy timelines of drug development—often costing billions of dollars and taking decades per approved drug—pose severe bottlenecks in addressing unmet medical needs. Conventional RL agents that maximize reward often generate promising molecular candidates that eventually fail essential safety checks, resulting in toxic or impractical compounds. Inspired by dual-process cognitive theory (Kahneman, 2011), our framework mirrors human decision-making by integrating: • System 1 (Intrinsic Curiosity): Drives rapid exploration by rewarding unexpected outcomes (Schmidhuber, 2010; Pathak et al., 2017). • System 2 (Reflective Reasoning): Preemptively validates candidate actions using rule-based constraints (e.g., Lipinski’s Rules, PAINS filters) combined with data-driven safety predictions
Author: Arjuna amrit, Vicky
Language: English
Subfield: Computational Theory and Mathematics
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
Funder: Latvijas Zinātnes Padome
OpenAlex ID: W6968645480
FWCI:
Cites:
Cited by:
Related to: