Explainable artificial intelligence evolves antimicrobial peptides
Related Articles
Comprehensive discovery and functional characterization of the noncanonical proteome
The systematic identification and functional characterization of noncanonical translation products, such as novel peptides, will facilitate the understanding of the human genome and provide new insights into cell biology. Here, we constructed a high-coverage peptide sequencing reference library with 11,668,944 open reading frames and employed an ultrafiltration tandem mass spectrometry assay to identify novel peptides. Through these methods, we discovered 8945 previously unannotated peptides from normal gastric tissues, gastric cancer tissues and cell lines, nearly half of which were derived from noncoding RNAs. Moreover, our CRISPR screening revealed that 1161 peptides are involved in tumor cell proliferation. The presence and physiological function of a subset of these peptides, selected based on screening scores, amino acid length, and various indicators, were verified through Flag-knockin and multiple other methods. To further characterize the potential regulatory mechanisms involved, we constructed a framework based on artificial intelligence structure prediction and peptide‒protein interaction network analysis for the top 100 candidates and revealed that these cancer-related peptides have diverse subcellular locations and participate in organelle-specific processes. Further investigation verified the interacting partners of pep1-nc-OLMALINC, pep5-nc-TRHDE-AS1, pep-nc-ZNF436-AS1 and pep2-nc-AC027045.3, and the functions of these peptides in mitochondrial complex assembly, energy metabolism, and cholesterol metabolism, respectively. We showed that pep5-nc-TRHDE-AS1 and pep2-nc-AC027045.3 had substantial impacts on tumor growth in xenograft models. Furthermore, the dysregulation of these four peptides is closely correlated with clinical prognosis. Taken together, our study provides a comprehensive characterization of the noncanonical proteome, and highlights critical roles of these previously unannotated peptides in cancer biology.
Cell-autonomous innate immunity by proteasome-derived defence peptides
For decades, antigen presentation on major histocompatibility complex class I for T cell-mediated immunity has been considered the primary function of proteasome-derived peptides1,2. However, whether the products of proteasomal degradation play additional parts in mounting immune responses remains unknown. Antimicrobial peptides serve as a first line of defence against invading pathogens before the adaptive immune system responds. Although the protective function of antimicrobial peptides across numerous tissues is well established, the cellular mechanisms underlying their generation are not fully understood. Here we uncover a role for proteasomes in the constitutive and bacterial-induced generation of defence peptides that impede bacterial growth both in vitro and in vivo by disrupting bacterial membranes. In silico prediction of proteome-wide proteasomal cleavage identified hundreds of thousands of potential proteasome-derived defence peptides with cationic properties that may be generated en route to degradation to act as a first line of defence. Furthermore, bacterial infection induces changes in proteasome composition and function, including PSME3 recruitment and increased tryptic-like cleavage, enhancing antimicrobial activity. Beyond providing mechanistic insights into the role of proteasomes in cell-autonomous innate immunity, our study suggests that proteasome-cleaved peptides may have previously overlooked functions downstream of degradation. From a translational standpoint, identifying proteasome-derived defence peptides could provide an untapped source of natural antibiotics for biotechnological applications and therapeutic interventions in infectious diseases and immunocompromised conditions.
Discovery of new AMR drugs targeting modulators of antimicrobial activity using in vivo silkworm screening systems
Global concerns about drug-resistant bacteria have underscored the need for new antimicrobial drugs. Emerging strategies in drug discovery include considering the third factors that influence drug activity. These factors include host-derived elements, adjuvants, and drug combinations, which are crucial in regulating antimicrobial efficacy. Traditional in vivo assessments have relied on animal models to study drug absorption, distribution, metabolism, excretion, and toxicity (ADMET). Alternative models, such as silkworms, are being explored to overcome the ethical and financial barriers associated with mammalian models. The silkworm has been proven effective in evaluating ADMET and in highlighting the therapeutic potential enhanced by third factors. Host factors (either mammalian or non-mammalian) enhance the antimicrobial activity of antimicrobial agents such as lysocin E. Additionally, using d-cycloserine to potentiate vancomycin has successfully combated vancomycin-resistant infections in silkworms. Leveraging silkworms in drug discovery could establish a novel screening method incorporating interactions with third factors, whether host related or non-host-related, thus promising new pathways for identifying antimicrobial drugs with unique mechanisms of action.
Towards next-gen smart manufacturing systems: the explainability revolution
The paper shares the author’s perspectives on the role of explainable-AI in the evolving landscape of AI-driven smart manufacturing decisions. First, critical perspectives on the reasons for the slow adoption of explainable-AI in manufacturing are shared, leading to a discussion on its role and relevance in inspiring scientific understanding and discoveries towards achieving complete autonomy. Finally, to standardize explainability quantification, a new Transparency–Cohesion–Comprehensibility (TCC) evaluation framework is proposed and demonstrated.
Deep learning enhances the prediction of HLA class I-presented CD8+ T cell epitopes in foreign pathogens
Accurate in silico determination of CD8+ T cell epitopes would greatly enhance T cell-based vaccine development, but current prediction models are not reliably successful. Here, motivated by recent successes applying machine learning to complex biology, we curated a dataset of 651,237 unique human leukocyte antigen class I (HLA-I) ligands and developed MUNIS, a deep learning model that identifies peptides presented by HLA-I alleles. MUNIS shows improved performance compared with existing models in predicting peptide presentation and CD8+ T cell epitope immunodominance hierarchies. Moreover, application of MUNIS to proteins from Epstein–Barr virus led to successful identification of both established and novel HLA-I epitopes which were experimentally validated by in vitro HLA-I-peptide stability and T cell immunogenicity assays. MUNIS performs comparably to an experimental stability assay in terms of immunogenicity prediction, suggesting that deep learning can reduce experimental burden and accelerate identification of CD8+ T cell epitopes for rapid T cell vaccine development.
Responses