MolPAL Active-Learning Virtual Screening

Overview

Plan or run MolPAL-style active-learning virtual screening from docking scores and candidate libraries. It is in the AI Spectra + Virtual Screening category and is intended to connect advanced spectra prediction, docking, and active-learning screening workflows.

When To Use It

• You need a focused workflow for molpal active-learning virtual screening without leaving ChemistryAtlas.
• You want a result that can be saved, shared, or chained into another chemistry app.
• You want the calculation assumptions and limitations visible next to the output.

Inputs

• text - Chemistry input - type: textarea - Use formulas, names, SMILES-like text, reactions, or key=value options. Heavier engines will plug into this same app surface.

Recommended Workflow

• Prepare molecules, spectra, proteins, or candidate libraries; run the sidecar-capable workflow; then inspect rankings, uncertainties, and failure diagnostics.
• Start with the smallest representative input, confirm the parser understood it, then scale to a larger list or workflow.
• Save the generated report when the result will feed a notebook entry, route review, model comparison, or team discussion.

Outputs

• A Markdown-style chemistry report with parsed inputs, assumptions, and calculated or predicted results.
• Structured tables when the app returns multiple compounds, reagents, routes, peaks, candidates, or model rows.
• Warnings, fallback notes, and sidecar availability messages when a specialized engine is not installed or not reachable.

Method And Backend Notes

This app has a runnable ChemistryAtlas backend path. Backend type: model. ChemistryAtlas roadmap MVP: runnable report now; specialist cheminformatics/model backend plugs into this app surface next. For production model use, keep ChemProp, MolPAL, and related engines in sidecar environments and record the model version, training set, and applicability domain.

How To Interpret Results

• Sidecar model quality depends on installation, model weights, preparation choices, and applicability domain; review raw files and controls before acting.
• Compare results across related molecules, controls, blanks, literature examples, or known reactions whenever possible.
• For decisions that affect safety, synthesis scale-up, biological testing, purchasing, or publication, verify with primary data and expert review.

Example Input

budget=2
smiles,score
CCO,-4.1
CCN,-4.5
c1ccccc1,
CCCl,

Common Checks Before Acting

• Confirm names, salts, stereochemistry, tautomers, protonation state, and hydration state.
• Check units, concentrations, equivalent definitions, and significant figures.
• Record external database versions, model versions, sidecar availability, and any manual edits made after the app output.

Related Apps

• AI MS/MS Predictor (ai-msms-predictor)
• Spectrum Similarity Scorer (spectrum-similarity-scorer)
• Candidate Structure Ranker (candidate-structure-ranker)
• Formula + Spectrum Elucidation (formula-spectrum-elucidation)
• PyScreener Docking Workflow (pyscreener-docking-workflow)

Acknowledgements And Validation

• ChemistryAtlas documentation and UI were prepared for chemistry discovery workflows.
• Where available, calculations may use open-source cheminformatics, reaction-informatics, spectra, docking, or machine-learning engines such as RDKit-family tooling, ASKCOS-style sidecars, ChemProp, ms-pred/ICEBERG, PyScreener, and MolPAL.
• Always verify important results against primary literature, official SDS records, instrument software, validated models, and local laboratory procedures.
• Model-driven outputs should include model version, training domain, uncertainty, and independent validation before operational use.