code for FVINN

During the dim 2020, an epidemic plagued human all over the world, which was caused by a notorious virus SARS-CoV-2. Since some foreseeing reports suppose this disaster to be long lasting, drugs to resist its further attack has been an urgent demand. While new drug discovery remains a long process. Exploring known moleculars may be an alternatively effective way to assist facing this challenge. In this study, we propose a fingerprint and virtual interaction neural network (FVINN) model for predicting drug-target affinity to assist screening drugs with high affinity to some target of SARS-CoV2, which may own the potentiality to deactivate SARS-Cov2. The drug fingerprint is calculated as circular fingerprints by RDkit. The protein fingerprint is generated by clustering patterns in protein sequence with distance of similarity for appearing together. The virtual interaction feature is generated by performing attentional mechanism on the one-dimension convoltional result of drug and protein sequence. The model is proved efficacious on two drugCtarget affinity benchmark datasets, KIBA and Davis datasets, with better Mean Squared Error(MSE),ConcordanceIndex(CI)and r2 thanDeepDTAand AttentionDTA. We finally give a case for this model to find some potential drugs with high affinity to target site Spike on SARSCoV-2, in which one has been confirmed potential for inhibiting the virus according to CHEMBL website.