(CBI2022 oral) REstretto: An efficient protein-ligand docking tool based on a fragment reuse strategy（CBI学会2022年大会）

Transcript

1. 25th October, 2022 CBI2022 O2-1 | K Yanagisawa 1 REstretto: An efficient protein-ligand docking tool based on a fragment reuse strategy 〇Keisuke Yanagisawa, Rikuto Kubota, Yasushi Yoshikawa, Masahito Ohue, Yutaka Akiyama School of Computing, Tokyo Institute of Technology 25th October, 2022 CBI Society Annual Meeting @ Funabori, Tokyo O2-1 O CH Cl HN NH CH3 Cl O O F OH O NH OH O
2. Abstract 25th October, 2022 CBI2022 O2-1 | K Yanagisawa 2 We developed “virtual screening-oriented” faster protein-ligand docking tool, REstretto https://github.com/akiyamalab/restretto • Fragment commonality enables acceleration • Comparable performance to AutoDock Vina • Further speed-up is promising O CH Cl HN NH CH3 Cl O O F OH O NH OH O K Yanagisawa, R Kubota et al., ACS Omega 7, 30265-30274, 2022.
3. Docking-based virtual screening 25th October, 2022 CBI2022 O2-1 | K Yanagisawa 3 Compound library Drug candidates Docking Filtering Screen candidates from a large compound library Limitations: • Docking is too slow to evaluate >10M compounds • Knowledge-based filtering decreased novelty → Faster docking tool can overcome the limitations
4. Reuse of common structures (fragments) 25th October, 2022 4 CBI2022 O2-1 | K Yanagisawa Existing docking tools evaluate compounds individually (without any reuse of intermediate results) Reuse of intermediate results will accelerate the docking calculation O CH Cl HN NH CH3 Cl O O F OH O NH OH O 0 100,000 200,000 The number of fragments A compound decomposition result for ZINC 12 Drugs Now Subset 10,639,555 cmpds x 1/50
5. Research Aim & Achievement 25th October, 2022 CBI2022 O2-1 | K Yanagisawa 5 Research Aim: Acceleration of docking tools for virtual screening • Assumption: single protein & huge compound library Achievement: Development of REstretto, a virtual screening-oriented docking tool • Accelerated via reuse of fragment docking results • Comparable performance to AutoDock Vina K Yanagisawa, R Kubota et al., ACS Omega 7, 30265-30274, 2022.
6. Overview of REstretto 25th October, 2022 CBI2022 O2-1 | K Yanagisawa 6 O CH Cl HN CH3 Cl O NH (B) Compound decomposition O CH Cl HN NH CH3 Cl O NH (C) Fragment grid generation (D) Rough conformer score calc. (E) Local Optimization (A) Conformer generation ⋮
7. Performance of REstretto 25th October, 2022 CBI2022 O2-1 | K Yanagisawa 7 Comparable performance to AutoDock Vina On execution time and virtual screening accuracy K Yanagisawa, R Kubota et al., ACS Omega 7, 30265-30274, 2022. Execution time [sec.] Accuracy (ROC-AUC)
8. Breakdown of execution time 25th October, 2022 CBI2022 O2-1 | K Yanagisawa 8 Grid generation (52.4%) and local optimization (35.1%) are the most time-consuming steps Average execution time per a compound with 10,000 compounds [sec.] • Grid generation is accelerated with larger library (>10,000) • Improvement of local optimization will accelerate more Proportional to #compound Proportional to #fragment K Yanagisawa, R Kubota et al., ACS Omega 7, 30265-30274, 2022.
9. Conclusion 25th October, 2022 CBI2022 O2-1 | K Yanagisawa 9 We developed “virtual screening-oriented” faster protein-ligand docking tool, REstretto • Accelerated with structural commonality • Comparable performance to AutoDock Vina • Further speed-up is promising with a library with >10,000 compounds O CH Cl HN NH CH3 Cl O O F OH O NH OH O K Yanagisawa, R Kubota et al., ACS Omega 7, 30265-30274, 2022.