Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Automated Genomic Interpretation via Concept Bottleneck Models for Medical Robotics

Zijun Li, Jinchang Zhang, Ming Zhang, Guoyu Lu

TL;DR

The paper tackles the challenge of translating genomic sequences into auditable, action-oriented decisions for medical robotics. It presents an end-to-end pipeline that maps DNA to CGR images, encodes them with a CNN, and enforces predictions through an interpretable concept bottleneck, augmented by priors, distribution alignment, and calibration. A cost-aware recommendation layer closes the loop from genomic interpretation to robotic decision policies, yielding interpretable evidence, improved calibration, and favorable decision economics. The approach achieves state-of-the-art classification on HIV gag datasets, demonstrates strong concept fidelity (GC, CpG, CCC) with high correlations and AUROCs, and shows robust faithfulness and decision-layer performance. Collectively, this work establishes a practical, auditable framework for integrating genomic interpretation into automated medical robotics and genomic medicine workflows.

Abstract

We propose an automated genomic interpretation module that transforms raw DNA sequences into actionable, interpretable decisions suitable for integration into medical automation and robotic systems. Our framework combines Chaos Game Representation (CGR) with a Concept Bottleneck Model (CBM), enforcing predictions to flow through biologically meaningful concepts such as GC content, CpG density, and k mer motifs. To enhance reliability, we incorporate concept fidelity supervision, prior consistency alignment, KL distribution matching, and uncertainty calibration. Beyond accurate classification of HIV subtypes across both in-house and LANL datasets, our module delivers interpretable evidence that can be directly validated against biological priors. A cost aware recommendation layer further translates predictive outputs into decision policies that balance accuracy, calibration, and clinical utility, reducing unnecessary retests and improving efficiency. Extensive experiments demonstrate that the proposed system achieves state of the art classification performance, superior concept prediction fidelity, and more favorable cost benefit trade-offs compared to existing baselines. By bridging the gap between interpretable genomic modeling and automated decision-making, this work establishes a reliable foundation for robotic and clinical automation in genomic medicine.

Automated Genomic Interpretation via Concept Bottleneck Models for Medical Robotics

TL;DR

The paper tackles the challenge of translating genomic sequences into auditable, action-oriented decisions for medical robotics. It presents an end-to-end pipeline that maps DNA to CGR images, encodes them with a CNN, and enforces predictions through an interpretable concept bottleneck, augmented by priors, distribution alignment, and calibration. A cost-aware recommendation layer closes the loop from genomic interpretation to robotic decision policies, yielding interpretable evidence, improved calibration, and favorable decision economics. The approach achieves state-of-the-art classification on HIV gag datasets, demonstrates strong concept fidelity (GC, CpG, CCC) with high correlations and AUROCs, and shows robust faithfulness and decision-layer performance. Collectively, this work establishes a practical, auditable framework for integrating genomic interpretation into automated medical robotics and genomic medicine workflows.

Abstract

We propose an automated genomic interpretation module that transforms raw DNA sequences into actionable, interpretable decisions suitable for integration into medical automation and robotic systems. Our framework combines Chaos Game Representation (CGR) with a Concept Bottleneck Model (CBM), enforcing predictions to flow through biologically meaningful concepts such as GC content, CpG density, and k mer motifs. To enhance reliability, we incorporate concept fidelity supervision, prior consistency alignment, KL distribution matching, and uncertainty calibration. Beyond accurate classification of HIV subtypes across both in-house and LANL datasets, our module delivers interpretable evidence that can be directly validated against biological priors. A cost aware recommendation layer further translates predictive outputs into decision policies that balance accuracy, calibration, and clinical utility, reducing unnecessary retests and improving efficiency. Extensive experiments demonstrate that the proposed system achieves state of the art classification performance, superior concept prediction fidelity, and more favorable cost benefit trade-offs compared to existing baselines. By bridging the gap between interpretable genomic modeling and automated decision-making, this work establishes a reliable foundation for robotic and clinical automation in genomic medicine.

Paper Structure

This paper contains 21 sections, 18 equations, 1 figure, 6 tables.

Table of Contents

  1. Introduction
  2. Relate Work
  3. Concept Bottleneck Models
  4. DNA representation learning
  5. Methodology
  6. Problem Definition
  7. Chaos Game Representation (CGR) Preprocessing
  8. Encoder and Concept Bottleneck
  9. Concept Supervision and Regularization
  10. Prior Consistency and Distribution Matching
  11. Uncertainty Estimation and Calibration
  12. Recommendation Layer: Cost-Aware Utility and Pairwise Ranking
  13. Joint Objective and Curriculum Schedule
  14. Experiment
  15. Datasets
  16. ...and 6 more sections

Figures (1)

  • Figure 1: Overall workflow of the proposed Automated DNA Parsing Framework. Raw DNA sequences are first transformed into Chaos Game Representation (CGR) images and processed by an image encoder to extract latent features. These features are mapped to a concept vector list through a concept bottleneck model, with prior consistency and concept fidelity constraints ensuring interpretability. The classifier outputs predictions, while a KL-divergence loss enforces consistency between concept-level predictions and feature-level predictions. Finally, the results are integrated into a cost-aware recommendation layer to accomplish an automated gene analysis process.