skills.rest
openai
openaiOFFICIAL
Explore all skills by this author →
WebsiteGitHub

boltz-small-molecule-screen

Screen SMILES compound libraries against protein targets using the Boltz API.

5.3k|765|Updated Mar 4, 2026
Search Tags:#virtual screening#drug discovery#molecular docking#smiles#boltz#adme#protein target
One-click install
npx skills add https://github.com/openai/plugins --skill boltz-small-molecule-screen
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: boltz-small-molecule-screen
Source: https://github.com/openai/plugins/tree/main/plugins/boltz-api-cli/skills/boltz-small-molecule-screen
Command: npx skills add https://github.com/openai/plugins --skill boltz-small-molecule-screen

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Screening a library of candidate molecules against a protein target normally requires docking infrastructure, scoring pipelines, and manual result wrangling. This Skill drives the Boltz small-molecule library-screen API end to end: normalizing SMILES inputs, estimating cost, submitting the job, downloading results, and ranking hits with free ADME triage data.

Core Features & Use Cases

  • Library normalization and payload authoring: Converts raw SMILES, CSV, .smi, or .txt compound lists plus protein sequences, pocket residues, and reference ligands into a valid screen payload.
  • Cost estimation and submission: Runs estimate-cost before start so users confirm the flat $0.025-per-molecule spend, with idempotency keys for safe re-runs.
  • Background download and ranking: Polls and downloads every per-hit structure, then ranks results by binding_confidence for hit discovery or optimization_score for lead optimization, including free Tier-1 ADME (solubility, permeability, lipophilicity) per molecule.
  • Use Case: A medicinal chemist has 500 candidate compounds and a kinase target sequence. The Skill builds the payload, confirms the $12.50 cost, submits the screen, downloads all predicted structures, and reports the top 10 binders with confidence metrics and ADME risk flags.

Quick Start

Use the boltz-small-molecule-screen skill to screen my compound library in candidates.csv against this protein sequence, estimate the cost first, and rank the top hits by binding confidence.

Frequently Asked Questions about boltz-small-molecule-screen

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I screen a compound library against a protein target with Boltz?▼

Normalize your SMILES library and target sequence into a payload with molecules and target fields, run estimate-cost to confirm spend, then submit with start and an idempotency key. Launch download-results in the background to poll and fetch every per-hit structure, then rank from results/index.jsonl.

How much does a Boltz small-molecule library screen cost?▼

Cost is a flat $0.025 per scored molecule, independent of molecule or target size. Always run the estimate-cost command before submitting, since pre-scoring molecule filters can reduce how many molecules are actually scored and lower the total.

What is the difference between binding_confidence and optimization_score?▼

binding_confidence is the primary ranking metric for hit discovery, finding any binder, while optimization_score ranks by binding strength for lead optimization. They are parallel intents, not a fallback hierarchy, so sort by whichever matches your goal.

Does the Boltz screen include ADME predictions?▼

Yes, every scored molecule returns a free Tier-1 ADME block with solubility (categorical), permeability (numeric), and lipophilicity (LogD). For ADME on bare SMILES without a target, use the boltz-small-molecule-adme skill instead.

Why are fewer results returned than molecules submitted?▼

Default server-side molecule_filters drop candidates before scoring, so the results directory count is often lower than the input count. Compute input IDs minus the external_id values in results/index.jsonl to identify dropped molecules, and check run.json for rejection summaries.

When should I not use the small-molecule library screen?▼

Do not use it for de novo molecule design, one-off docking of a single compound, or ADME prediction on bare SMILES with no target. It is designed for screening existing candidate libraries against a defined protein target.

Related Skills

Explore More Education & Research Skills→
adaptyvbioadaptyvbio
151

uniprot

Fetch protein sequences and annotations from the UniProt REST API.

Official
Basic
junior1pjunior1p
11

ipsae

Rank protein binder designs using ipSAE_min, LIS, and pDockQ scores.

Community
Intermediate
biomaps-infrabiomaps-infra

bio-diffdock

Predict molecular docking poses for protein-ligand interactions using diffusion models.

Official
Advanced
NVIDIA-BioNeMoNVIDIA-BioNeMo
413

drug-discovery-pipeline

Generate, dock, and score drug candidates via GenMol, DiffDock, and Boltz2 NIMs.

Official
Advanced