Additional file 1 of AMPlify: attentive deep learning model for discovery of novel antimicrobial peptides effective against WHO priority pathogens

Additional file 1: Supplementary Note S1: Performance comparison of AMP Scanner Vr.2 and AMPlify re-trained on the AMP Scanner Vr.2 dataset. Supplementary Note S2: Performance comparison of different AMP prediction tools based on the test sequence similarities to their corresponding training sets. Supplementary Note S3: Comparison of different AMP prediction tools tested on similar sequences with different labels. Supplementary Figure S1: Learning curve comparison (on the validation sets for early stopping) of single sub-models of AMPlify trained on two different datasets. (a) Single sub-models of AMPlify trained on our own training set; (b) Single sub-models of AMPlify trained on the AMP Scanner Vr.2 “Train+Tune” partitions. Square markers denote the best epochs chosen by early stopping, and the x-axes have been set in the same range in order for a clearer comparison. Supplementary Figure S2: UpSet plot of the 101 candidate mature sequences with regard to the three filters. This plot visualizes the results obtained by applying different combinations of filters to the 101 candidate mature sequences. Supplementary Figure S3: Workflow of selecting the non-AMP sequences from the UniProtKB/Swiss-Prot database. Supplementary Figure S4: Workflow of the AMP discovery pipeline. The process describes how 75 putative AMPs were identified from the bullfrog genome. Invalid sequences denote those not suitable for AMPlify prediction, with lengths outside the range 2 to 200 amino acids or with non-standard amino acids. Supplementary Table S1: Stratified 5-fold cross-validation results of different architectures on the training set. The top section compares the architecture of AMPlify, with and without ensemble learning, with its simpler variations. The second section shows the architecture of AMP Scanner Vr.2 cross-validated on our training set. Values of accuracy (acc), sensitivity (sens), specificity (spec), F1 score (F1) and area under the receiver operating characteristic curve (AUROC) are presented along with their standard deviations in percentage. Supplementary Table S2: Comparison between AMP Scanner Vr.2 and AMPlify cross-validated on all data provided by AMP Scanner Vr.2 (“Train+Tune+Test” partitions). This table shows the 10-fold cross-validation results of AMP Scanner Vr.2 and AMPlify on all data provided by AMP Scanner Vr.2. Values of accuracy (acc), sensitivity (sens), specificity (spec) and area under the receiver operating characteristic curve (AUROC) are presented along with their standard deviations in percentage. Supplementary Table S3: Performance comparison between AMP Scanner Vr.2 and AMPlify re-trained on the AMP Scanner Vr.2 “Train+Tune” partitions and tested on their “Test” partition. Since AMPlify applies early stopping and the exact size of training set for each single sub-model is smaller, the exact training size for each model is listed here in the second column. Values of accuracy (acc), sensitivity (sens), specificity (spec) and area under the receiver operating characteristic curve (AUROC) are presented in percentage. Supplementary Table S4: Minimum inhibitory concentrations (MIC) and minimum bactericidal concentrations (MBC) of selected AMP candidates following antimicrobial susceptibility testing (AST) in vitro. This is a supplementary table to Table 3. Candidate antimicrobial peptides were synthesized and purchased from Genscript. AST, and MIC/MBC determination was performed as outlined by the Clinical and Laboratory Standards Institute (CLSI), with modification as recommended by Hancock. Data is presented as the lowest effective peptide concentration range (μg/mL) observed in three independent experiments. LL37, human cathelicidin and a peptide from Tp0751 from Treponema pallidum were used as the positive and negative control peptides, respectively. Supplementary Table S5: Predictions of Gaegurin 5 (GGN5) and its analogues by different AMP prediction tools. Antimicrobial activity data of GGN5 and its analogues were taken from the work by Won and con-workers. The analogues were generated by truncating the parent peptide into shorter fragments and/or by amino acid substitutions. Prediction results of AMPlify, AMP Scanner Vr.2, and iAMPpred were listed for comparison.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.