Alignment-Free Deep Learning Framework Discovers Antimicrobial Peptides in Insect Proteomes

Researchers have developed iAMP-Deep, an alignment-free, multilayered artificial intelligence framework designed to discover insect antimicrobial peptides without relying on traditional sequence alignment. Insect antimicrobial peptides represent a central component of innate immunity and serve as promising templates for developing next-generation antimicrobials. However, established identification techniques rely heavily on sequence alignment and motif similarity, which can overlook divergent, short, or noncanonical peptides. Furthermore, existing insect antimicrobial peptide databases remain taxonomically limited, and transcriptomic screening pipelines frequently yield large volumes of speculative candidates. The new framework was engineered to address these limitations by combining physicochemical descriptors, motif and domain signatures, and protein-language model embeddings with ensemble learning and anomaly-aware post-processing.

To train and evaluate the system, the researchers compiled a curated dataset of experimentally validated insect antimicrobial peptides alongside length-matched non-antimicrobial peptides, documenting their source accession numbers and family labels. To reduce homology-driven leakage between data partitions, sequence clustering at eighty percent identity or greater was conducted prior to data splitting. The team then trained four complementary machine learning architectures: LightGBM, deep neural networks, deep convolutional neural networks, and transformer neural networks, using fixed random seeds and documented hyperparameters.

Across individual model evaluations, the deep convolutional neural network demonstrated the strongest standalone performance, reaching an area under the receiver operating characteristic curve of 0.998. The final ensemble model achieved a precision of 0.97, a recall of 0.96, and an F1 score of 0.96, misclassifying only six antimicrobial peptides and 23 non-antimicrobial peptides during testing. Physicochemical profiling confirmed that insect antimicrobial peptides are enriched in cationic, amphipathic, and moderately hydrophobic residues, forming a distinct feature space relative to non-antimicrobial peptides. Prediction confidence was further refined by applying post-processing stages using Isolation Forest algorithms alongside antimicrobial-peptide-specific physicochemical constraints.

The researchers applied iAMP-Deep to the proteome of the giant honeybee, Apis dorsata, identifying 15 high-confidence candidates that contained 13 unique peptide sequences. These predicted peptides exhibited characteristic antimicrobial peptide properties, including positive net charges ranging from +2.50 to +12.06, hydrophobicity ratios between 0.302 and 0.500, amphipathicity scores between 1.321 and 2.243, and instability index values below 45. The candidates also displayed conserved signatures corresponding to cecropin-, defensin-, and melittin-like motifs, and functional enrichment analysis indicated that 82 percent of the candidates were associated with immune defense and secretory pathways.

The candidate set recovered from Apis dorsata included both motif-positive peptides and motif-poor peptides that maintained strong physicochemical and protein-language model support despite low global similarity to reference sequences, confirming the framework’s capacity to discover candidates beyond canonical motifs. In evaluations on a balanced independent test set, the complete anomaly-filtered pipeline achieved a precision of 0.959, a recall of 0.942, and an F1 score of 0.950, whereas conventional baseline methods achieved F1 scores ranging from 0.012 to 0.667.

To evaluate broader taxonomic utility, the unaltered pipeline was applied to 13 proteomes across four additional insect orders: Hemiptera, Lepidoptera, Diptera, and Coleoptera. This external screen yielded 416 high-confidence entries representing 237 unique peptide sequences. Combined with the initial Apis dorsata case study in Hymenoptera, the evaluations spanned five insect orders and consistently recovered noncanonical, lysine-rich antimicrobial-peptide-like candidates with high ensemble model confidence.

By integrating curated insect antimicrobial peptide data with physicochemical descriptors, motif and domain signatures, protein-language model embeddings, ensemble learning, and anomaly filtering, iAMP-Deep provides an interpretable and alignment-free discovery pipeline. Compared with motif-only and alignment-based baselines, the framework is designed to retain detection sensitivity for divergent candidates while reducing false positives and prioritizing peptides for downstream experimental testing.

Subject of Research: Computational discovery of insect antimicrobial peptides using artificial intelligence

Article Title: iAMP-deep: an alignment-free AI framework for insect antimicrobial peptide discovery

Article References: Kolanchi, P., Marimuthu, M., Thulasy, S., & Pagalahalli Sankaran, S. (2026). iAMP-deep: an alignment-free AI framework for insect antimicrobial peptide discovery. BMC Genomics. https://doi.org/10.1186/s12864-026-13439-7

Image Credits: AI Generated

DOI: 10.1186/s12864-026-13439-7

Keywords: antimicrobial peptides, iAMP-Deep, computational biology, deep learning, protein language models, insect immunity, proteomics, ensemble learning, alignment-free, framework, insect, antimicrobial