Deep-Learning Model for Tumor-Type Prediction Using Targeted Clinical Genomic Sequencing Data

Madison Darmofal ^1,2, Shalabh Suman ³, Gurnit Atwal ^4,5,6

¹Computational and Systems Biology Program, Sloan Kettering Institute, Memorial Sloan Kettering Cancer Center, New York, New York.
²Tri-Institutional Training Program in Computational Biology and Medicine, Weill Cornell Medicine, New York, New York.
³Department of Pathology, Memorial Sloan Kettering Cancer Center, New York, New York.
⁴Computational Biology Program, Ontario Institute for Cancer Research, Toronto, Ontario, Canada.
⁵Department of Molecular Genetics, University of Toronto, Toronto, Ontario, Canada.
⁶Vector Institute, Toronto, Ontario, Canada.
⁷Department of Medicine, Memorial Sloan Kettering Cancer Center, New York, New York.
⁸Marie-Josée and Henry R. Kravis Center for Molecular Oncology, Memorial Sloan Kettering Cancer Center, New York, New York.
⁹Human Oncology and Pathogenesis Program, Memorial Sloan Kettering Cancer Center, New York, New York.
¹⁰Department of Epidemiology and Biostatistics, Memorial Sloan Kettering Cancer Center, New York, New York.

⁰Computational and Systems Biology Program, Sloan Kettering Institute, Memorial Sloan Kettering Cancer Center, New York, New York.

Cancer Discovery +

|

February 28, 2024

View abstract on PubMed

Summary

A new deep learning model, Genome-Derived-Diagnosis Ensemble (GDD-ENS), accurately predicts tumor type using targeted gene panel sequencing. This approach rivals whole-genome sequencing methods and aids in diagnosing rare cancers for improved patient treatment.

Area Of Science

Oncology
Genomics
Bioinformatics

Background

Histology-based cancer diagnosis is challenging, yet tumor type is crucial for treatment decisions.
Genomic alterations are highly diagnostic of tumor type, but current methods are often not clinically feasible.
Existing tumor-type classifiers face limitations with whole-genome sequencing data or restricted cancer type predictions.

Purpose Of The Study

To develop a clinically feasible and accurate tumor-type classification model using targeted cancer gene panel sequencing data.
To create a deep neural network-based hyperparameter ensemble for enhanced diagnostic performance.
To enable real-time tumor-type predictions for guiding clinical treatment decisions.

Main Methods

Utilized genomic features from 39,787 solid tumors sequenced via a targeted cancer gene panel.
Developed Genome-Derived-Diagnosis Ensemble (GDD-ENS), a deep neural network ensemble model.
Trained and validated the model for classifying tumor types across 38 distinct cancer categories.

Main Results

GDD-ENS achieved 93% accuracy for high-confidence predictions across 38 cancer types.
The model's performance rivals that of whole-genome sequencing-based methods.
GDD-ENS demonstrated utility in diagnosing rare tumor types and cancers of unknown primary, with potential for incorporating clinical information.

Conclusions

GDD-ENS provides a clinically feasible and accurate method for tumor-type prediction using targeted gene panels.
The model's performance rivals more complex whole-genome sequencing approaches.
Integrating GDD-ENS into clinical workflows can enhance real-time diagnostic capabilities and guide cancer treatment decisions.