A Deep Learning Approach to Antibiotic Discovery

(Cell 180, 688–702.e1–e13; February 20, 2020) Our paper reported the use of a machine learning approach to discover new antibacterial molecules. Since publication, we have become aware of the following errors in our paper that we are now correcting. (1) The structure of ZINC000100032716, shown in the Graphical Abstract and Figure 6D, mistakenly displayed a carbonyl carbon making five bonds. We have redrawn the molecules to display the correct structures. (2) In Figure S3A and the accompanying legend, the concentrations of halicin used were 20 µg/mL (10x MIC) and 40 µg/mL (20x MIC), not 10 µg/mL and 20 µg/mL. (3) In Figure S5K, the x axis labels should range from 10-6 to 103, in accordance with the axis tick marks, not 10-5 to 103. (4) In Table S2B, BRD-K57502136-345-03-4, BRD-K90177246-001-05-5, BRD-K15514357-001-05-6, and BRD-A56621826-001-02-1 were listed as being unavailable for empirical validation. However, the four molecules that were not available for testing were BRD-K76819217-001-01-4, BRD-A41063939-001-01-0, BRD-M10279501-065-05-9, and BRD-A40472231-304-02-5. This resulted from a transposition error in converting the original training data file into the Table S2B spreadsheet. This transposition error was not present in model training. (5) In the “Initial model training and the identification of halicin” section of the Results, the halicin prediction ranks noted in parentheses should read “positions ranging from 273 to 2579,” rather than “positions raging from 273 to 1987.” (6) In the “Bacterial cell killing assays” subsection of the STAR Methods, the M. tuberculosis strain used was H37Rv, not “M37Rv.” (7) In the “Mutant generation” section, ΔnfsA::kan was mistakenly written as “ΔnsfA::kan.” (8) The section title “baumannii mouse infection model” should have been “A. baumannii mouse infection model.” (9) It was brought to our attention that SU3327 (which we renamed halicin) had been reported as an active compound in an unpublished screen, deposited to PubChem, for growth inhibition of M. tuberculosis. The following sentence has been added to the last paragraph of the Results section “Halicin is a broad-spectrum bactericidal antibiotic” to acknowledge this: “The molecule we have named halicin was reported to have growth inhibitory activity against M. tuberculosis in a high-throughput screening setting (unpublished data; PubChem AID 1259343).” (10) In preparing the final version of the manuscript, we inadvertently misspelled the last name of author Zohar Bloom-Ackermann as ‘‘Zohar Bloom-Ackerman.” These errors have now been corrected in the online version of the paper. We apologize for any inconvenience they may have caused the readers.Figure 6. Predicting New Antibiotic Candidates from Unprecedented Chemical Libraries (original)View Large Image Figure ViewerDownload Hi-res image Download (PPT)Figure S3. Mechanistic Investigations into Halicin, Related to Figure 4 (Corrected)View Large Image Figure ViewerDownload Hi-res image Download (PPT)Figure S3. Mechanistic Investigations into Halicin, Related to Figure 4 (original)View Large Image Figure ViewerDownload Hi-res image Download (PPT)Figure S5. Model Predictions from the WuXi Anti-tuberculosis Library and the ZINC15 Database, Related to Figure 6 (Corrected)View Large Image Figure ViewerDownload Hi-res image Download (PPT)Figure S5. Model Predictions from the WuXi Anti-tuberculosis Library and the ZINC15 Database, Related to Figure 6 (original)View Large Image Figure ViewerDownload Hi-res image Download (PPT)Graphical Abstract (corrected)View Large Image Figure ViewerDownload Hi-res image Download (PPT)Graphical Abstract (original)View Large Image Figure ViewerDownload Hi-res image Download (PPT) A Deep Learning Approach to Antibiotic DiscoveryStokes et al.CellFebruary 20, 2020In BriefA trained deep neural network predicts antibiotic activity in molecules that are structurally different from known antibiotics, among which Halicin exhibits efficacy against broad-spectrum bacterial infections in mice. Full-Text PDF Open Archive