DrugForge-ADMET v0.3.0: An auditable BRICS-based workflow
- 1Department of Biology
- 2Çukurova University
Abstract
Early hit discovery normally begins with a physical or virtual collection of small molecules. Public resources such as PubChem, ChEMBL and ZINC have broadened access to chemical space, but using them effectively still requires retrieval, standardization, filtering, deduplication, three-dimensional preparation and provenance management; many laboratories also lack an in-house screening collection. We present DrugForge-ADMET v0.3.0, an open-source RDKit workflow intended to provide a compact, auditable starting set when a prepared library is unavailable. The software decomposes 30 inspectable, sp3-enriched seed templates into BRICS fragments, recombines them, sanitizes the products, and applies deterministic physicochemical rules, structural-alert catalogues and explicitly labelled descriptor-based risk proxies. Seed/reference similarity is assessed with radius-2, 2048-bit Morgan fingerprints and Tanimoto similarity; score-first diversity selection limits redundancy. Selected structures are exported as CSV and as SDF records with ETKDGv3 conformers followed, where parameterized, by MMFF or UFF optimization. In three fixed-seed technical validation runs, each capped at 10,000 enumerated products, 61, 157 and 194 candidates were selected. Across 412 outputs, 411 canonical SMILES were unique, all 412 SDF records were readable and all contained three-dimensional conformers. The pooled median values were molecular weight 355.50, cLogP 1.55, TPSA 79.25 Å2, fraction sp3 0.706, three rotatable bonds, one aromatic ring and QED 0.845. These results establish software execution and property-window compliance, not bioactivity, synthesizability, patent novelty or experimentally validated ADMET performance. DrugForge-ADMET is therefore positioned as a transparent candidate-generation and triage layer before target-specific modelling, retrosynthesis and experimental testing.
Related
- Integrating dft-based spectroscopic analysis, molecular docking, admet prediction, and natural biosorbent adsorption studies of flupyradifurone
- AuditRxn: Auditable Fast-Weight Memories for Multi-Step Chemical Reaction Prefix-Prediction
- Sliding Window Attention with Auditable Chemical Sinks for Long-Sequence Molecular Transformation Modeling
- Accelerating Lead Optimization away from Anti-Targets: Navigating the Structural Complexity of Promiscuous ADMET Proteins
- ADMET property prediction and toxicity profiling of CDK2 inhibitors: a practical cheminformatics tutorial using ADMET-AI 2.0 and RDKit
BibTeX
@misc{istifli2026drugforgeadmet,
title = {{DrugForge-ADMET v0.3.0: An auditable BRICS-based workflow}},
author = {Erman Salih İSTİFLİ},
howpublished = {ChemRxiv},
year = {2026},
month = oct,
doi = {10.26434/chemrxiv.15009708/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15009708/v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - DrugForge-ADMET v0.3.0: An auditable BRICS-based workflow
AU - Erman Salih İSTİFLİ
PY - 2026
DA - 2026/10/01
PB - ChemRxiv
DO - 10.26434/chemrxiv.15009708/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15009708/v1
AB - Early hit discovery normally begins with a physical or virtual collection of small molecules. Public resources such as PubChem, ChEMBL and ZINC have broadened access to chemical space, but using them effectively still requires retrieval, standardization, filtering, deduplication, three-dimensional preparation and provenance management; many laboratories also lack an in-house screening collection. We present DrugForge-ADMET v0.3.0, an open-source RDKit workflow intended to provide a compact, auditable starting set when a prepared library is unavailable. The software decomposes 30 inspectable, sp3-enriched seed templates into BRICS fragments, recombines them, sanitizes the products, and applies deterministic physicochemical rules, structural-alert catalogues and explicitly labelled descriptor-based risk proxies. Seed/reference similarity is assessed with radius-2, 2048-bit Morgan fingerprints and Tanimoto similarity; score-first diversity selection limits redundancy. Selected structures are exported as CSV and as SDF records with ETKDGv3 conformers followed, where parameterized, by MMFF or UFF optimization. In three fixed-seed technical validation runs, each capped at 10,000 enumerated products, 61, 157 and 194 candidates were selected. Across 412 outputs, 411 canonical SMILES were unique, all 412 SDF records were readable and all contained three-dimensional conformers. The pooled median values were molecular weight 355.50, cLogP 1.55, TPSA 79.25 Å2, fraction sp3 0.706, three rotatable bonds, one aromatic ring and QED 0.845. These results establish software execution and property-window compliance, not bioactivity, synthesizability, patent novelty or experimentally validated ADMET performance. DrugForge-ADMET is therefore positioned as a transparent candidate-generation and triage layer before target-specific modelling, retrosynthesis and experimental testing.
ER -