OsteoToxPred Bone Toxicity Prediction Platform

Platform Overview

Drug‑induced bone toxicity refers to adverse effects of certain compounds on bone metabolism, density and structure. Existing toxicity models rarely focus specifically on bone toxicity. We propose a multimodal method combining molecular fingerprints and graph‑based representations, deployed as the online platform OsteoToxPred. It achieves strong performance on bone toxicity prediction (Accuracy 0.85, AUC 0.92), outperforming multiple ML baselines. As an efficient, interpretable and easy‑to‑use tool, OsteoToxPred supports safety assessment and research, helping scientists make informed decisions.

System and Model Architecture

Diagram: Fingerprint/GNN dual branches, gated fusion and serving pipeline

Use Cases

Early Safety Screening

Rapidly identify potential bone‑toxic candidates to reduce R&D costs.

Mechanism Studies

Leverage contributing features/substructures to hypothesize and verify mechanisms.

Structure Optimization

Use highlighted fragments to guide de‑risking modifications.

Environmental Health

In‑vitro prioritization for regulatory decision support.

Core Technical Highlights

BTP-MFFGNN Multimodal Model multimodal feature fusion Graph Neural Networks (GNN) Attention Mechanism drug development Batch Prediction Visualization

Start Predicting View Help

Key Features

Multimodal Prediction Engine (BTP-MFFGNN)

Fingerprints + graph representations; attention and gated fusion for fine‑grained recognition. Acc: 0.85, AUC: 0.92.

Visualized Results

Return labels (toxic/non‑toxic) and probabilities, with molecule images for easy comparison.

Batch Inference

SMILES‑based batch prediction; one per line, up to 50 per run, with sample filler.

Explainability

LIME‑based local explanations highlight contributing features/substructures.

Quick Start

Experience OsteoToxPred now