Reinforcement learning to boost molecular docking upon protein conformational ensemble.

Area of Science:

• Biochemistry
• Computational Biology
• Drug Discovery

Background:

• Intrinsically disordered proteins (IDPs) are implicated in various human diseases, making them important therapeutic targets.
• The conformational flexibility of IDPs presents a significant computational challenge for traditional drug screening methods.
• Docking large ligand libraries to numerous IDP conformations is computationally prohibitive.

Purpose of the Study:

• To develop an efficient computational method for virtual screening of intrinsically disordered proteins (IDPs).
• To address the challenge posed by the conformational ensemble of IDPs in drug discovery.
• To optimize the docking process for large ligand libraries against IDPs.

Main Methods:

• Proposal of a reversible upper confidence bound (UCB) algorithm tailored for IDP virtual screening.
• Dynamic arrangement of the docking process to focus computational effort.
• Application of the algorithm to the transcription factor c-Myc as a case study.

Main Results:

• The reversible UCB algorithm significantly reduces the average docking computation per ligand.
• The screening performance is minimally impacted despite the reduction in computational effort.
• Demonstrated efficiency in accurately separating top-ranking ligands from the bulk.

Conclusions:

• Reinforcement learning, specifically the UCB algorithm, offers a highly efficient solution for IDP virtual screening.
• This approach effectively overcomes the computational bottleneck associated with IDP conformational ensembles.
• The method holds promise for advancing rational drug design for diseases involving IDPs.