MLMarker: a machine learning framework for tissue inference and biomarker discovery
Clinical Snapshot
PICO Framework
| P — Population | Proteomics datasets derived from human tissues, cancer cohorts, and biofluids (including cerebral melanoma metastases, pan-cancer samples, and biofluid specimens) |
| I — Intervention | MLMarker — a Random Forest-based machine learning framework trained on 34 healthy human tissues, generating continuous tissue similarity scores with SHAP-based protein-level explanations and a missing-protein penalty factor |
| C — Comparator | No explicit head-to-head comparator tool described; performance evaluated against known tissue-of-origin labels in public datasets and internal validation cohorts |
| O — Outcomes | Accuracy of tissue-of-origin inference, interpretability of protein-level biomarker contributions (SHAP values), robustness in low-coverage/sparse proteomics datasets, and hypothesis generation for cancer biology |
Bottom Line
MLMarker is a novel Random Forest-based computational framework that generates continuous tissue similarity scores from proteomics data, with SHAP-based protein-level interpretability and a penalty mechanism for sparse datasets. Trained on 34 healthy human tissues and validated retrospectively across three public datasets — including cerebral melanoma metastases, a pan-cancer cohort, and biofluids — it demonstrates conceptual promise for tissue-of-origin inference and biomarker hypothesis generation. However, the evidence base at this stage is preliminary. Critical performance metrics (AUC, sensitivity, specificity, confidence intervals) are absent from the abstract, no prospective clinical validation has been conducted, and the tool has not been benchmarked against existing tissue-of-origin classifiers. The training on healthy tissue proteomes may limit performance in pathological specimens. For Australian clinicians, the tool is not yet ready for clinical diagnostic use but represents a potentially valuable research instrument for CUP investigation and translational proteomics. Prospective validation in clinically annotated cohorts, with rigorous reporting of diagnostic accuracy metrics and clinical utility endpoints, is required before integration into pathology workflows. Researchers with proteomics capabilities should monitor this space as the tool matures.
Key Findings
P Value: Not reported
Effect Size: Not quantitatively reported in the abstract; described qualitatively as 'high accuracy' for the pan-cancer cohort
Primary Outcome: Tissue-of-origin inference accuracy across three independent public proteomics datasets: (1) brain-like protein signatures identified in cerebral melanoma metastases; (2) high classification accuracy reported in a pan-cancer cohort; (3) brain and pituitary tissue origins inferred from biofluid proteomics data
Nnt Or Sensitivity: Sensitivity, specificity, AUC, and other diagnostic accuracy metrics are not reported in the abstract. For a tissue inference tool, these would be the critical performance parameters. SHAP values provide protein-level feature importance scores but are not equivalent to diagnostic accuracy metrics.
Confidence Interval: Not reported
Clinical Application
The tool is available as an open-source Python package and Streamlit web application, lowering the technical barrier for research use. However, clinical implementation requires: (1) standardised proteomics sample preparation pipelines; (2) bioinformatics expertise for data pre-processing; (3) integration with laboratory information management systems; (4) prospective clinical validation before diagnostic use. Current feasibility is primarily in research and hypothesis-generation settings rather than routine clinical diagnostics. In Australia, cancer of unknown primary (CUP) affects approximately 2–5% of all cancer diagnoses annually, representing a significant clinical challenge where tissue-of-origin inference tools could add value. The TGA does not currently regulate bioinformatics software tools as medical devices unless they meet the definition of an IVD medical device under the Therapeutic Goods (Medical Devices) Regulations 2002 — clinical deployment of MLMarker for diagnostic purposes would require regulatory consideration. The RCPA (Royal College of Pathologists of Australasia) and RACGP guidelines for CUP management currently rely on immunohistochemistry, gene expression profiling, and next-generation sequencing; proteomics-based tissue inference is not yet embedded in Australian standard-of-care pathways. Medicare Benefits Schedule (MBS) does not currently list proteomics-based tissue inference as a reimbursable diagnostic service. Research adoption through Australian Genomics or ACRF-funded centres is the most plausible near-term pathway. Potentially applicable to patients with cancer of unknown primary (CUP), ambiguous metastatic lesions, or complex biofluid proteomics profiles where tissue-of-origin is diagnostically uncertain. Also relevant for translational research contexts including biomarker discovery in oncology, neurology, and endocrinology.
Abstract
MLMarker is a machine learning tool that computes continuous tissue similarity scores for proteomics data, addressing the challenge of interpreting complex or sparse datasets. Trained on 34 healthy tissues, its Random Forest model generates probabilistic predictions with SHAP-based protein-level explanations. A penalty factor corrects for missing proteins, improving robustness for low-coverage samples. Across three public datasets, MLMarker revealed brain-like signatures in cerebral melanoma metastases, achieved high accuracy in a pan-cancer cohort, and identified brain and pituitary origins in biofluids. MLMarker provides an interpretable framework for tissue inference and hypothesis generation, available as a Python package and Streamlit app.
References
- 1.Claeys, T., van Puyenbroeck, S., Gevaert, K., & Martens, L. (2026). MLMarker: a machine learning framework for tissue inference and biomarker discovery. Genome Biology. https://doi.org/10.1074/mcp.M800231-MCP200
Related Research
The American journal of pathology
An Integrated Machine Learning and Genomic Framework for Precise Detection of Gastric Cancer
3 Aug 2026
Current opinion in chemical biology
Prediction of protein-protein interactions and co-complex models with deep learning
3 Aug 2026
European journal of radiology
A CT-based deep learning model to differentiate between benign and malignant adrenal lesions
3 Aug 2026
This content is for educational purposes for healthcare professionals only and does not constitute clinical advice. Clinical decisions should be based on individual patient assessment, current guidelines, and appropriate specialist consultation. Editorial Standards · Privacy Policy · Terms of Service