Submission Details
Molecule(s):
CCC(C1N=CC=C1CO)C(O)/C(C1=C(CO)NC(=O)CC1)=C(\C)Cl
CCNC1=NC=CCC1CC1(C2(O)CC(F)=CCO2)CCN(C)C1
CCCNC1=NCC(C#N)C=C1CC1C(C)NC2=NC=CCC21
CNC(=O)N[C@H]1CCOC(N)(CO)O1
CC(CCO)(C1=C(O)COC(O)=C1)[C@H]1CC[C@@H](CO)O1
C[C@H](C1=C(O)CC(O)=C1)C1(C(=O)O)CC=C(O)CC1C(=O)O
CNC(=O)NC1=C([C@H](C)C(=O)NC)CC(O)C1
N#CC1=CCCC=C1C1=NC(SCC(=O)NC2=CCN=CC2C(F)F)=NC(O)C1C#N
Design Rationale:
These designs are based on a generative machine learning model trained on very broad collection of known anti-viral compounds (HIV-1 Protease inhibitors, Ebola and Herpes antivirals, ...) and a hand curated selection of related molecules. The set of molecules generated by this model are then scored on chemical similarity to the relevant conformers of Umifenovir (in Covid-19 trials as a single agent), Vonoprazan (highest scoring compound in the recent FOLDING@HOME docking run), and Galidesivir (known for broad-spectrum antiviral activity). The resulting high-scoring molecules were filtered (softly) on a series of AMDE and toxicity conditions, including Veber, Egan and Muegge violations. Most molecules presented here are P-glycoprotein substrates and have a high GI absorption, and all have a favorable bioavailability score. The final selection of molecules was decided on by hand based on a range of factors including synthetic accessibility (ASKCOS), toxicity (SwissADME) and some diversity measures.
Other Notes:
Special thanks to Jeriek Van den Abeele at UiO's Department of Physics for computational support and to ASKCOS and SwissADME for their publicly accessible servers.