PreditX®: an end-to-end machine learning webserver for target-specific virtual screening and ADMET-guided hit prioritization

Machine learning-based virtual screening (MLVS) is increasingly used to prioritize chemical matter across large compound libraries, yet practical adoption remains limited by fragmented workflows, dependence on curated datasets, evaluation leakage, and limited support for deployable target-specific prediction. We developed PreditX®, a cloud-deployed MLVS webserver that automates target-level bioactivity retrieval, curation, activity labeling, scaffold-aware model training, external prediction, consensus ranking, and absorption, distribution, metabolism, excretion, and toxicity (ADMET)-supported hit triage. PreditX® integrates UniProt/ChEMBL-based data retrieval, standardized filtering, adaptive pActivity-based class assignment, Bemis–Murcko scaffold splitting, Morgan fingerprints, Mordred descriptors, multi-model training, and downloadable prediction reports. We benchmarked PreditX® across 105 data-rich targets, of which 104 produced complete performance outputs. Across evaluable targets, median best area under the receiver operating characteristic curve (ROC-AUC) was 0.9557, with 88/104 targets exceeding 0.90 and 61/104 exceeding 0.95. Random Forest was the most consistently competitive model, achieving the highest ROC-AUC for 74/104 targets. Cross-target analysis showed that activity-label separability, captured by the pActivity threshold margin, was the strongest observed correlate of performance, while novelty and scaffold-diversity metrics identified difficult generalization regimes. As a case study, epidermal growth factor receptor (EGFR) models trained from automatically curated ChEMBL data were used for consensus prediction over a purchasable multi-million-compound library, returning a ranked, ADMET-annotated candidate set. EGFR sensitivity analyses preserved discrimination under endpoint restriction but revealed moderate ranking changes, while 4,605 withheld intermediate compounds showed graded prediction behavior. PreditX® provides an integrated, reproducible, and interpretable MLVS workflow for early-stage compound prioritization. PreditX® webserver is accessible at: https://app.preditx.com/home.

Authors

Publication Details

Journal
Bioinformatics Advances
Published
2026-11-19
DOI
https://doi.org/10.1093/bioadv/vbag306
Primary Topic
Computational Drug Discovery Methods
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

PreditX®: an end-to-end machine learning webserver for target-specific virtual screening and ADMET-guided hit prioritization

Kanzal Iman, Muhammad Usman Mirza, Jorge Emilio Alfonso
Bioinformatics Advances
Computational Drug Discovery Methods
article

PreditX®: an end-to-end machine learning webserver for target-specific virtual screening and ADMET-guided hit prioritization

Kanzal Iman, Muhammad Usman Mirza, Jorge Emilio Alfonso
article en

Abstract

Machine learning-based virtual screening (MLVS) is increasingly used to prioritize chemical matter across large compound libraries, yet practical adoption remains limited by fragmented workflows, dependence on curated datasets, evaluation leakage, and limited support for deployable target-specific prediction. We developed PreditX®, a cloud-deployed MLVS webserver that automates target-level bioactivity retrieval, curation, activity labeling, scaffold-aware model training, external prediction, consensus ranking, and absorption, distribution, metabolism, excretion, and toxicity (ADMET)-supported hit triage. PreditX® integrates UniProt/ChEMBL-based data retrieval, standardized filtering, adaptive pActivity-based class assignment, Bemis–Murcko scaffold splitting, Morgan fingerprints, Mordred descriptors, multi-model training, and downloadable prediction reports. We benchmarked PreditX® across 105 data-rich targets, of which 104 produced complete performance outputs. Across evaluable targets, median best area under the receiver operating characteristic curve (ROC-AUC) was 0.9557, with 88/104 targets exceeding 0.90 and 61/104 exceeding 0.95. Random Forest was the most consistently competitive model, achieving the highest ROC-AUC for 74/104 targets. Cross-target analysis showed that activity-label separability, captured by the pActivity threshold margin, was the strongest observed correlate of performance, while novelty and scaffold-diversity metrics identified difficult generalization regimes. As a case study, epidermal growth factor receptor (EGFR) models trained from automatically curated ChEMBL data were used for consensus prediction over a purchasable multi-million-compound library, returning a ranked, ADMET-annotated candidate set. EGFR sensitivity analyses preserved discrimination under endpoint restriction but revealed moderate ranking changes, while 4,605 withheld intermediate compounds showed graded prediction behavior. PreditX® provides an integrated, reproducible, and interpretable MLVS workflow for early-stage compound prioritization. PreditX® webserver is accessible at: https://app.preditx.com/home.

Bioinformatics Advances
Computational Drug Discovery Methods
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon