Table 1 Comparison of different pipeline setups

Setup	Target	Ranking	Active learning	Compounds computationally screened	Total simulation time (h)	Compounds to screen experimentally	rs
Baseline	Receptor ensemble	Docking score	Yes	2755.2	*15,612.8	1299.4	0.4
No MD scoring	Receptor ensemble	Static h-score	Yes	262.4	*1486.9	5.6	0.2
Full active learning approach	Receptor ensemble	Dynamic h-score	Yes	246.0	*3403.0	7.6	1.0
No receptor ensemble	Homology model	Dynamic h-score	Yes	754.4	829.8	709.0	0.0
No receptor ensemble nor MD scoring	Homology model	Static h-score	Yes	2230.4	631.9	2179.8	0.6
No active learning cycle	Receptor ensemble	Dynamic h-score	No	7166.8	*99,140.7	16.6	1.0

Target structure, ranking method, usage of active learning, number of computationally screened compounds, total simulation (docking + molecular dynamics) time, number of compounds to screen experimentally and Spearman’s coefficient for rank correlation of four known inhibitors for different pipeline setups. Average (mean) values over five replicates are reported. Total simulation time is expressed in core hours, which indicates one CPU/GPU being used for one hour of execution time. Total simulation times denoted by * additionally include the time required to generate the receptor ensemble, which we estimate to be ~8000 h. Note that Spearman’s coefficient was calculated using only four known inhibitors and this limited sample size may influence the statistical reliability of the correlation. The best results in each category are marked in bold.

Quick links

Search