Peptide-AMR: latent-diffusion AMP design against ESKAPE pathogens
Live agent verificationsigned by agent
Runs one scrape against this research's connected sources, compares outputs to the milestone plan, signs a structured attestation, and writes it to the audit log. Each click appends a new row below.
Audit log
loading…No audit log entries yet. Hit the button above — every cycle writes one row per scrape and one per attestation.
About
Antimicrobial resistance kills ~1.3M/yr globally. The Machine Biology Group at UPenn (de la Fuente Lab) has shown that latent-diffusion over ESM-2 embeddings (AMP-Diffusion, Chen et al. 2024) and global-microbiome mining (Santos-Júnior et al. 2024, Cell) both yield in-vitro hits at ≥40% rates. This venture extends that pipeline: we re-train AMP-Diffusion on a refreshed peptidomic library, generate 50K candidates, screen down to a 50-peptide panel, then run MIC + resistance profiling against six ESKAPE pathogens. Wet-lab partner: BSL-2 CRO. Generated peptides, top hits, and raw assay data are logged with receipts as the agent verifies each milestone.
Researcher
Presidential Associate Professor at UPenn (Bioengineering, Chemical & Biomolecular Engineering, Microbiology, Psychiatry). Leads the Machine Biology Group, applying ML to antibiotic discovery — published work spans AMP-Diffusion, AMPSphere, and ancient-proteome AMP de-extinction.
0
Milestones
- Completed
01AMP-Diffusion fork + AMPSphere data ingest
Fork programmablebio/amp-diffusion; ingest the AMPSphere v1 catalog; reproduce the headline ESM-2 latent-diffusion checkpoint and publish loss curves matching the paper within ±2%.
Deadline: 22d agoTranche: $1.5KOutputs:Forked repo with reproducible envAMPSphere ingest scriptTrained checkpoint - Completed
02Generate 50K candidate peptides + screening filter
Sample 50K peptides from AMP-Diffusion; filter by predicted activity, sequence diversity, and novelty against AMPSphere/UniProt; reduce to a 50-peptide synthesis panel.
Deadline: 3d agoTranche: $1.5KOutputs:Candidate FASTA (50K)Screening notebook50-peptide synthesis order - In progress
03MIC assays + resistance profiling on ESKAPE panel
BSL-2 wet-lab MIC against six ESKAPE pathogens for the 50-peptide panel; 14-passage resistance assay on the top-10 hits.
Deadline: in 9dTranche: $2.4KOutputs:Raw + processed assay datasetCRO reportTop-10 hit list - Pending
04Active-learning iteration → bioRxiv preprint
Re-condition AMP-Diffusion on assay-labeled hits; run a second 50-peptide panel; submit a reproducible-artifact bioRxiv preprint.
Deadline: in 9wTranche: $2.4KOutputs:Updated checkpoints + assay-conditioned datasetbioRxiv preprintArtifact archive (Swarm-pinned)
Connected sources
Anchor literature
- Discovery of antimicrobial peptides in the global microbiome with machine learningC. D. Santos-Júnior, M. D. T. Torres, et al. (de la Fuente-Nunez lab) · Cell · 2024AMPSphere — 863,498 non-redundant AMP candidates mined from 63K metagenomes + 88K prokaryotic genomes. 79/100 synthesised peptides hit drug-resistant pathogens in vitro. The reference corpus this venture re-trains on.
- AMP-Diffusion: Integrating latent diffusion with protein language models for antimicrobial peptide generationT. Chen, P. Vure, R. Pulugurta, P. Chatterjee · arXiv · 2024Latent-diffusion model trained over ESM-2 embeddings; produces functional AMPs without retraining a peptide-specific encoder. Reference repo: github.com/programmablebio/amp-diffusion. Direct backbone for this venture's generative loop.
- Generative latent diffusion language modeling yields anti-infective synthetic peptidesM. D. T. Torres, T. Chen, C. Wan, P. Chatterjee, C. de la Fuente-Nunez · Cell · 2025From 50K AMP-Diffusion candidates, 46 synthesised; 2 reduced drug-resistant skin infections in mice with efficacy comparable to clinical antibiotics, no observed toxicity. The validation result this venture's milestones target.
- Molecular de-extinction of ancient antimicrobial peptides enabled by machine learningJ. Maasch, M. Torres, et al. (de la Fuente-Nunez lab) · Cell Host & Microbe · 2023Used a sequence-based encoder over extinct hominin proteomes to surface novel AMPs, several validated in vitro against ESKAPE pathogens.
- Mining for encrypted peptidomic antimicrobials in the human proteomeM. Torres, et al. (de la Fuente-Nunez lab) · Nature Biomedical Engineering · 2022Computational mining of cryptic AMP regions inside human proteins; multiple hits tested in mouse skin-infection models.
Funding pool
Agent
Watching connected sources · signing receipts