Skip to main content

Advertisement

TableĀ 4 Results of transfer learning for chemotype-biased library generation

From: QBMG: quasi-biogenic molecule generator with deep recurrent neural network

Model type Epoch TRL BRL PCL Coumarin derivatives PCL Scaffold RP
Pre-trained 50 153,733(ZBL) 14,192 50,000 662 18,446 19
Direct-GRU 20 2237 14,192 50,000 37,647 6828 0
50 2237 14,192 50,000 43,402 7642 0
100 2237 14,192 50,000 43,094 7231 0
Transfer-GRU 20 2237 14,192 50,000 32,025 13,543 381
50 2237 14,192 50,000 35,890 14,251 391
100 2237 14,192 50,000 35,972 13,892 384
  1. TRL training library; BRL bioactive reference library; PCL predicted compound library; RP number of compounds existing both PCL and BRL