REINVENT4 and RDKit
Generate SMILES, canonicalise structures, remove invalid records, and filter by Lipinski, Veber, PAINS, scaffold, similarity, and Gram-negative entry heuristics.
PROTO-NOOS: Orchestrating Open-Access Bioinformatics for Seamless Drug Discovery
A reproducible PROTO-NOOS pipeline for prioritising small-molecule candidates in an E. coli DHFR context by combining molecular generation, structural prediction, short-MD checks, systems biology, cell target engagement, and retrosynthetic accessibility.
Abstract
Computational antibiotic discovery requires combining chemical generation, structural modelling, physical validation, biological interpretation, and synthetic feasibility into a reproducible workflow. This study presents the PROTO-NOOS pipeline, an in silico pipeline for prioritising small-molecule candidates in an E. coli DHFR context. The workflow used REINVENT4 for de novo molecule generation, RDKit-based physicochemical and Gram-negative entry filtering, Boltz2 for protein-ligand complex and affinity prediction, GROMACS for molecular dynamics stability checks, BioTransformer3 and COBRApy/iML1515 for metabolite and target-perturbation analysis, CellTE for ODE-based intracellular target engagement, and AiZynthFinder for retrosynthetic accessibility. The pipeline connects chemical, structural, MD, metabolic, target-level, and retrosynthetic evidence in one screening process. Its outputs should be interpreted as computational ranking features for hypothesis generation, not as confirmed antibacterial activity, because synthesis, antibacterial assays, toxicity testing, and wet-lab validation were outside the scope of this work.
100de novo candidates passed into the representative downstream run
27non-zero PROTO-NOOS ranked candidates after multi-criteria scoring
2.3 hobserved runtime for the 100-compound pipeline run
6/10top-ranked candidates with solved retrosynthetic routes in the reported batch
Pipeline Design
PROTO-NOOS treats early antibiotic discovery as a staged prioritisation problem. Each tool contributes a weak but inspectable signal, and the final ranking is based on agreement across chemical, structural, biophysical, cellular, and synthetic constraints rather than on a single affinity score.
Boltz2 and GROMACS
Predict protein-ligand complexes for PDB 6XG5, estimate affinity-related signals, and run short MD gates for physical plausibility.
BioTransformer and COBRApy
Map metabolite predictions into iML1515, simulate target-level folA perturbation, and separate metabolic context from direct binding evidence.
CellTE and AiZynthFinder
Model intracellular target occupancy with entry, efflux, association, and dissociation terms, then screen ranked candidates for retrosynthetic feasibility.
Main Findings
The current run demonstrates workflow feasibility and exposes where the evidence is strong, weak, or contradictory. The key value is structured triage: fewer candidates move into expensive downstream PROTO-NOOS BCA, where pocket contacts, NADPH context, MM-GBSA, PLUMED metadynamics, and selected QM/DFT descriptors can be inspected in greater detail.
Multi-signal ranking narrows the search space.
Of 100 generated compounds, 27 retained a non-zero PROTO-NOOS score; the rest were treated as resistant or low-priority under the current evidence model.
CellTE adds context beyond KD.
Post-Boltz2 rescoring combines predicted affinity with entry and efflux dynamics, allowing compounds with similar KD values to separate by Cout50 and AUCbound.
Short MD remains a low-confidence gate.
The 100 ps high-throughput mode is useful for flagging unstable systems, but kinetic proxies are explicitly treated as weak labels.
Retrosynthesis changes lead priority.
The top PROTO-NOOS-ranked molecule was not solved by AiZynthFinder in the reported batch, while several lower-ranked candidates had practical synthetic routes.
Cell target engagement scoring
ODE-based intracellular target occupancy reframes affinity as one part of a cellular accumulation and binding process.
Figures
The figures below summarise the evidence blocks used by the manuscript: branch-level PROTO-NOOS distributions, inter-stage agreement, CellTE rescore behaviour, retrosynthetic feasibility, and final target-engagement interpretation.
Cross-branch score distributions
Shows how different evidence branches shape candidate prioritisation before final interpretation.
Inter-stage signal agreement
Spearman correlations identify which pipeline stages support each other and where evidence diverges.
Pre-score versus rescore
Compares CellTE ranking before and after Boltz2-derived KD_pred is introduced into the occupancy model.
Retrosynthetic accessibility
Contrasts predicted biological priority with route solvability, route score, stock coverage, and route depth.
Poster and Manuscript
The poster gives the compact visual version of the workflow, while the manuscript describes the full scientific scope, limitations, stage contracts, and interpretation rules.
Scope and Limits
PROTO-NOOS is an in silico prioritisation framework, not a confirmed antibiotic discovery. The current study does not include compound synthesis, MIC testing, target-engagement assays, cytotoxicity assays, toxicity validation, or wet-lab antibacterial experiments. The practical conclusion is therefore limited to workflow feasibility, hypothesis generation, and candidate triage before deeper simulation or experimental follow-up.
Selected References
- Stokes, J. M., Yang, K., Swanson, K., et al. A deep learning approach to antibiotic discovery. Cell, 2020.
- Abraham, M. J., Murtola, T., Schulz, R., Pall, S., Smith, J. C., Hess, B., and Lindahl, E. GROMACS: High performance molecular simulations through multi-level parallelism. SoftwareX, 2015.
- Passaro, S., Corso, G., Wohlwend, J., et al. Boltz-2: Towards accurate and efficient binding affinity prediction. bioRxiv, 2025.
- Djoumbou-Feunang, Y., Fiamoncini, J., Gil-de-la-Fuente, A., et al. BioTransformer: a comprehensive computational tool for small molecule metabolism prediction. Journal of Cheminformatics, 2019.
- Loeffler, H. H., He, J., Tibo, A., et al. Reinvent 4: Modern AI-driven generative molecule design. Journal of Cheminformatics, 2024.
- Genheden, S., Thakkar, A., Chadimova, V., et al. AiZynthFinder: a fast, robust and flexible open-source software for retrosynthetic planning. Journal of Cheminformatics, 2020.
- Drusano, G. L. Pharmacokinetics and pharmacodynamics of antimicrobials. Clinical Infectious Diseases, 2007.