Individual and collective human intelligence in drug design: evaluating the search strategy

Table 6 Comparison of human individual and collective design experiments with automatic de novo designers reported in the GuacaMol publication [43]

Complexity level	Benchmark type	Final score Top-1 score (Top-10 score) (Top-100 score)
Complexity level	Benchmark type	Human individual	Human collective	smiles_lstm	graph_ga	smiles_ga	graph_mcts	best_of_chembl
L2	Rediscovery	0.72	1.0	1.0	1.0	0.732	0.355	0.505
L2	Rediscovery	0.72	1.0	1.0	1.0	0.732	0.355	0.505
L3	Rediscovery	0.61	1.0	1.0	1.0	0.598	0.311	0.456
L3	Rediscovery	0.61	1.0	1.0	1.0	0.598	0.311	0.456
L5	Rediscovery	0.80	1.0	1.0	1.0	0.515	0.311	0.419
L5	Rediscovery	0.80	1.0	1.0	1.0	0.515	0.311	0.419
L1	Similarity	0.99	1.0	1.0	1.0	0.907	0.749	0.719
		1.0	1.0	1.0	1.0	1.0	0.80	0.765
		1.0	1.0	1.0	1.0	1.0	0.758	0.726
		0.96	1.0	1.0	1.0	0.72	0.689	0.664
L4	Similarity	1.0	1.0	1.0	1.0	0.834	0.380	0.595
		1.0	1.0	1.0	1.0	0.856	0.428	0.609
		1.0	1.0	1.0	1.0	0.838	0.376	0.601
		1.0	1.0	1.0	1.0	0.807	0.335	0.576

Benchmark scores are explained in section "Comparison with automatic de novo designers". The final score is equivalent to the top-1 score in rediscovery benchmarks and to the average of top-1, top-10 and top-100 scores in the similarity ones

ISSN: 1758-2946