Deep learning-based design of novel antimicrobial peptides.

a. The overall framework of GAN generator and AMPredictor.

b. Visualization of GAN training process via UMAP and t-SNE. The newly generated sequences fit the distribution of known AMP space gradually.

c. Global alignment match scores between the generated sequences and itself and AMPs in the training set.

d. Six physiochemistry properties of generated peptides and real AMP sets.

e. Regression results of AMPredictor model on its test set.

f. The sequences of three selected peptides as well as their predicted and experimentally validated MIC values against E. coli ATCC 25922.

Antibacterial and toxicological evaluations of P076 peptide.

a. In vitro antibacterial assessment of peptides by determination of MICs.

b. Hemolysis of increasing concentrations of peptides. Results are shown as the means ± standard deviation (SDs).

c. Survival curves of mice given P076 and PB. Arrows indicate the drug concentration inducing a 50% death.

d. H&E staining of mouse kidneys. The scale bar indicates 200 μm.

e. Survival of infected mice treated with P076 and CIP.

f. g. Bacterial colonization of MDRAB in mouse peritoneal lavage fluid and spleen. Results are shown as the means ± SDs. ***, P<0.001. n.s., no significance.

Antiviral assays of AMPs against CHIKV, HTNV, DENV-2, and HSV-1.

a. Quantitative real-time PCR of virus RNA with gradient concentrations of P001, P002, and P076.

b. Immunofluorescence results of four viruses with gradient concentrations of P002.

All experiments are performed in triplicates. Column bars are means ± SEMs. * P<0.05, ** P<0.01, *** P<0.001, **** P<0.0001. ns, no significance.

Transmission electron microscopy (TEM) of viral infection with P002.

a. b. Cells infected with virus, where the red arrows mark the viral particles.

c. Cells are treated with 10 μM P002 and infected with virus (MOI=1), where no viral proliferation is observed.

d. Control of intact cells with 10 μM P002. All experiments are performed in triplicates.

Bacterial membrane-acting assessment of AMPs.

a. Binding kinetics of three AMPs to lipid A measured by biolayer interferometry.

b. Binding thicknesses of 500 nM AMPs to immobilized LPS within 120 seconds.

c. Fluorescence intensity of NPN after excitation at 350 nm. The maximum emission appeared at approximately 396 nm, where P001, P002, and P076 were detected higher intensity than control.

d. The distance between COM of peptides and phosphorus atoms of lipid A (around the midplane of G-bacterial outer membranes) during the last 50 ns of the MD simulation trajectories. The gray dashed line displays the initial distance at 0 ns.

e. Snapshots of P076 with G-bacterial outer membranes at 0 ns, 50 ns, and 100 ns. Only the top leaflet (LPS) is shown. Tan balls represent the phosphorus atoms of lipid A molecules.