Fig. 5: Schematic of the antimicrobial operation mechanism of AMP and the evaluation of generated AMP sequences.

a AMPs bursting the bacteria epidermis by pore formation. b An illustration of the transformation of an input sequence into a 20-dimensional vector (or 20 channels) with 24-time steps, where 20 is the number of the unique amino-acids in the input sequence with the added end cursor “@”, and 24 is the length of the sequence after adding the end cursor. c The percentage of AMP sequences in the generation set from short-term (low W/I ratio), long-term (high W/I ratio), and mixed case (various W/I ratios), respectively. d Average global charge from each sequence set. The mixed case shows a similar global charge value to the training set, while the others do not. Inset: the total global charge value of the entire sequences in each set. e Average molecular weight from each sequence set. The mixed case generates sequences that have similar weights to the training set. Inset: the total molecular weights of all of the sequences. f Amino-acid distribution of each sequence set. g The 3D image of an AMP (protein data bank ID: 2mag⁴⁴) in the training set. h A predicted 3D image of the generated AMP from the mixed case neuro-memristive computing system.