Analysis of Mutations in Precision Oncology using The Automated, Accurate, and User-Friendly Web Tool PredictONCO

Rayyan Tariq Khan, Petra Pokorná, Jan Štourač, Simeon Borko, Adam Dobias, Joan Planas-Iglesias, Stanislav Mazurenko, Ihor Arefiev, Gaspar Pinto, Veronika Szotkowská, Jaroslav Štěrba, Jiřı́ Damborský, Ondřej Slabý, David Bednář

Abstract

Next-generation sequencing technology has created many new opportunities for clinical diagnostics, but it faces the challenge of functional annotation of identified mutations. Various algorithms have been developed to predict the impact of missense variants that influence oncogenic drivers. However, computational pipelines that handle biological data must integrate multiple software tools, which can add complexity and hinder non-specialist users from accessing the pipeline. Here, we have developed an online user-friendly web server tool PredictONCO that is fully automated and has a low barrier to access. The tool models the structure of the mutant protein in the first step. Next, it calculates the protein stability change, pocket level information, evolutionary conservation, and changes in ionisation of catalytic amino acid residues, and uses them as the features in the machine-learning predictor. The XGBoost-based predictor was validated on an independent subset of held-out data, demonstrating areas under the receiver operating characteristic curve (ROC) of 0.95 and 0.94, and the average precision from the precision-recall curve 0.98 and 0.94 for structure-based and sequence-based predictions, respectively. Finally, PredictONCO calculates the docking results of small molecules approved by regulatory authorities. We demonstrate the applicability of the tool by presenting its usage for variants in two cancer-associated proteins, cellular tumour antigen p53 and fibroblast growth factor receptor FGFR1. Our free web tool will assist with the interpretation of data from next-generation sequencing and navigate treatment strategies in clinical oncology: https://loschmidt.chemi.muni.cz/predictonco/ .

Introduction

Next-Generation Sequencing (NGS) has revolutionized clinical diagnostics by enabling the identification of thousands of genetic mutations in cancer patients. However, the primary bottleneck remains "functional annotation"—distinguishing between "driver" mutations that promote oncogenesis and "passenger" mutations that are biologically neutral. While numerous algorithms exist to predict variant effects, they often require manual integration of disparate software tools, creating a significant barrier for clinical oncologists and non-specialist researchers. PredictONCO addresses this challenge by providing a fully automated, "one-stop-shop" web server that streamlines the process from structural modeling to therapeutic suggestion.

Methods

The PredictONCO pipeline utilizes a multi-step automated process to evaluate missense mutations.

User Interface and Input

The tool is designed for ease of use, requiring minimal input such as the protein name and the specific mutation. Users can select target proteins from a database and input mutations either textually or via a sequence table.

PredictONCO user interface showing protein selection (A), textual mutation input (B), and sequence table selection (C).

Computational Pipeline

  1. Structural Modeling: The tool first generates a 3D model of the mutant protein.
  2. Feature Extraction: Several biophysical properties are calculated, including protein stability changes ($\Delta\Delta G$) using FoldX and Rosetta, pocket analysis via P2Rank, evolutionary conservation, and ionization changes using PropKa.
  3. Machine Learning Prediction: These features are processed by an XGBoost-based predictor to generate an "oncogenicity score."
  4. Drug Docking: AutoDock Vina simulates the binding of over 4,000 FDA/EMA-approved small molecules to identify potential therapeutic candidates or resistance markers.
3D structural model of a protein used for feature extraction and pocket analysis in PredictONCO.

Datasets and Validation

The predictor was trained on a curated dataset of 464 oncogenic and 449 non-oncogenic mutations sourced from ClinVar and OncoKB. Validation was performed on an independent "held-out" subset to ensure robust performance.

Results

PredictONCO demonstrates high predictive accuracy and clinical relevance.

  • Performance Metrics: The structure-based predictor achieved an Area Under the ROC Curve (AUC) of 0.95 and an average precision of 0.98, indicating a very low false-positive rate.
  • Case Studies: The tool successfully identified the well-known p53 R175H inactivating variant. It also provided a high-confidence (98%) deleterious prediction for the less-characterized K139M variant.
  • Efficiency: For complex proteins like FGFR1, calculations were completed within approximately 6 hours, fitting within standard clinical diagnostic timelines.

Discussion

PredictONCO represents a significant step toward democratizing bioinformatics in oncology. By prioritizing 3D structural data, the tool provides higher confidence predictions for rare variants compared to sequence-only methods.

However, some limitations remain. High server demand can occasionally extend processing times. Currently, the tool is optimized for missense mutations and small molecule docking. Future developments aim to include complex alterations (e.g., large indels, translocations) and expand docking capabilities to include biologics like monoclonal antibodies. Prospective clinical validation will be essential to confirm the tool's impact on real-time patient outcomes.

Web tool availability: https://loschmidt.chemi.muni.cz/predictonco/

Summary updated

Abstract

Abstract Next-generation sequencing technology has created many new opportunities for clinical diagnostics, but it faces the challenge of functional annotation of identified mutations. Various algorithms have been developed to predict the impact of missense variants that influence oncogenic drivers. However, computational pipelines that handle biological data must integrate multiple software tools, which can add complexity and hinder non-specialist users from accessing the pipeline. Here, we have developed an online user-friendly web server tool PredictONCO that is fully automated and has a low barrier to access. The tool models the structure of the mutant protein in the first step. Next, it calculates the protein stability change, pocket level information, evolutionary conservation, and changes in ionisation of catalytic amino acid residues, and uses them as the features in the machine-learning predictor. The XGBoost-based predictor was validated on an independent subset of held-out data, demonstrating areas under the receiver operating characteristic curve (ROC) of 0.95 and 0.94, and the average precision from the precision-recall curve 0.98 and 0.94 for structure-based and sequence-based predictions, respectively. Finally, PredictONCO calculates the docking results of small molecules approved by regulatory authorities. We demonstrate the applicability of the tool by presenting its usage for variants in two cancer-associated proteins, cellular tumour antigen p53 and fibroblast growth factor receptor FGFR1. Our free web tool will assist with the interpretation of data from next-generation sequencing and navigate treatment strategies in clinical oncology: https://loschmidt.chemi.muni.cz/predictonco/ .
BibTeX
@inproceedings{khan2024analysis,
  author = {Rayyan Tariq Khan and Petra Pokorná and Jan Štourač and Simeon Borko and Adam Dobias and Joan Planas-Iglesias and Stanislav Mazurenko and Ihor Arefiev and Gaspar Pinto and Veronika Szotkowská and Jaroslav Štěrba and Jiřı́ Damborský and Ondřej Slabý and David Bednář},
  title = {{Analysis of Mutations in Precision Oncology using The Automated, Accurate, and User-Friendly Web Tool PredictONCO}},
  booktitle = {Computational},
  year = {2024},
  doi = {10.1016/j.csbj.2024.11.026},
}
Figure