Explainable Machine Learning to Identify Patient-specific Biomarkers for Lung Cancer

Sobhan, M.; Mondal, A. M.

doi:10.1101/2022.10.13.512119

Home/Archive/bioRxiv

bioRxiv · 10.1101/2022.10.13.512119

Explainable Machine Learning to Identify Patient-specific Biomarkers for Lung Cancer

Sobhan, M.·Mondal, A. M.

View original source ↗Export metadata ↓

Abstract

BackgroundLung cancer is the leading cause of death compared to other cancers in the USA. The overall survival rate of lung cancer is not satisfactory even though there are cutting-edge treatment methods for cancers. Genomic profiling and biomarker gene identification of lung cancer patients may play a role in the therapeutics of lung cancer patients. The biomarker genes identified by most of the existing methods (statistical and machine learning based) belong to the whole cohort or population. That is why different people with the same disease get the same kind of treatment, but results in different outcomes in terms of success and side effects. So, the identification of biomarker genes for individual patients is very crucial for finding efficacious therapeutics leading to precision medicine. MethodsIn this study, we propose a pipeline to identify lung cancer class-specific and patient-specific key genes which may help formulate effective therapies for lung cancer patients. We have used expression profiles of two types of lung cancers, lung adenocarcinoma (LUAD) and lung squamous cell carcinoma (LUSC), and Healthy lung tissues to identify LUAD- and LUSC-specific (class-specific) and individual patient-specific key genes using an explainable machine learning approach, SHaphley Additive ExPlanations (SHAP). This approach provides scores for each of the genes for individual patients which tells us the attribution of each feature (gene) for each sample (patient). ResultIn this study, we applied two variations of SHAP - tree explainer and gradient explainer for which tree-based classifier, XGBoost, and deep learning-based classifier, convolutional neural network (CNN) were used as classification algorithms, respectively. Our results showed that the proposed approach successfully identified class-specific (LUAD, LUSC, and Healthy) and patient-specific key genes based on the SHAP scores. ConclusionThis study demonstrated a pipeline to identify cohort-based and patient-specific biomarker genes by incorporating an explainable machine learning technique, SHAP. The patient-specific genes identified using SHAP scores may provide biological and clinical insights into the patients diagnosis.

Explore related subjects

cancer biology Preprint

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Sobhan, M., Mondal, A. M.. 2022-10-17. Explainable Machine Learning to Identify Patient-specific Biomarkers for Lung Cancer. https://doi.org/10.1101/2022.10.13.512119

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

BAF complexes cooperate with lineage-restricted transcription factors to establish cell-state-specific regulatory programs

BRG1/BRM-associated factor (BAF) complexes are essential regulators of chromatin organization and cell identity. Genes encoding BAF subunits are among the most frequently mutated in human cancer. Yet, how disruption of a single subunit gives rise to distinct tumor states remains poorly understood. Atypical teratoid rhabdoid tumors (ATRTs) provide a compelling model, as they are uniformly driven by bi-allelic loss of the BAF subunit SMARCB1, yet segregate into distinct molecular subtypes. Using integrated multi-omics profiling of patient-derived ATRT organoids and primary tumor tissues, we show that transcriptional and chromatin accessibility changes upon SMARCB1 reconstitution are directed by subtype- and lineage-specific transcription factor (TF) repertoires. Genome-wide binding and transcriptomic analyses reveal that SMARCB1 reconstitution enables the lineage-specific TF RFX4 to engage regulatory elements and activate neuronal differentiation programs, partially reprogramming ATRT-MYC toward an ATRT-SHH-like identity. Together, these findings demonstrate that SMARCB1 loss disrupts TF-dependent chromatin priming, revealing that the lineage-specific regulatory output of BAF complexes is constrained by the underlying TF repertoire.

cancer biology↗

bioRxiv2026

Curvature-based schedule ordering is preserved in homogeneous and heterogeneous tumors

During treatment, dosing protocols with identical total dose may favor one schedule over another depending on how exposure is distributed over time. In homogeneous cell populations, Jensen's inequality has been shown to predict whether continuous or intermittent high-dose followed by low-dose is favored, based on the curvature of the concentration-effect function. Scheduling ordering can be performed based on curvature only: concave functions favor continuous dosing while convex functions favor intermittent dosing schedules with a high time-variance. However, tumors exhibit heterogeneous drug sensitivities, and treatment preferentially eliminates drug-sensitive cells, progressively shifting the surviving population toward greater drug resistance. Whether this treatment-induced selection can reverse curvature-based schedule ordering remains unclear. We develop a mathematical framework in which phenotypes share a common concentration-effect function but differ by fixed, nonnegative sensitivity scales. We prove that the ordering of equal-mean schedules is administrated solely by the scalar cumulative treatment effect, that is, by the curvature of the effect function alone. Consequently, baseline heterogeneity and treatment-induced selection can alter the magnitude of treatment response and the composition of surviving cells, but cannot reverse the ordering of equal-mean schedules predicted by the curvature of the single-cell concentration-effect function. This preservation holds across different temporal dosing profiles and repeated treatment cycles. Our work uncovers a general principle that curvature-based schedule ordering is preserved in heterogeneous tumors despite treatment-induced selection toward resistance.

cancer biology↗

bioRxiv2026

Enhanced antitumour activity of combined PD-1 blockade and HER2-targeted therapy is associated with reduced epithelial-mesenchymal transition and extracellular matrix-remodelling signatures in gastric cancer

Background The combination of PD-1 blockade, HER2-targeted therapy and chemotherapy improves outcomes in HER2-positive advanced gastric and gastro-oesophageal junction adenocarcinoma. However, chemotherapy can independently alter the tumour microenvironment, making the biological effects of PD-1 and HER2 targeting difficult to distinguish. We therefore investigated how PD-1 and HER2 blockade, alone and in combination, affect tumour growth and associated transcriptional programmes. Methods Immunodeficient NOG mice received human PBMCs by tail-vein injection. Three days later, NCI-N87 gastric cancer cells were implanted subcutaneously. Mice were randomised to vehicle, pembrolizumab, trastuzumab, or combination treatment. Tumour growth was assessed, and bulk RNA sequencing was used to identify treatment-associated changes in gene expression, pathway enrichment, and tumour, stromal, and immune cell-associated signatures. Results Combined treatment produced greater tumour growth control than either monotherapy. Trastuzumab-treated tumours showed enrichment of immune activation, epithelial-mesenchymal transition (EMT), and TGF{beta}-associated signatures, whereas pembrolizumab alone produced a limited transcriptional effect. Combination treatment was associated with reduced proliferation, EMT, and stromal-remodelling signatures, alongside increased cytotoxic immune-associated signatures. Expression of extracellular matrix-associated genes, including VCAN, SPARC, POSTN, collagen genes, and FAP, was lower in combination-treated tumours. Complement-related transcriptional pathways were also enriched. Conclusions In this chemotherapy-free model, combined PD-1 blockade and HER2-targeted therapy was associated with greater tumour growth control, enrichment of cytotoxic immune-associated transcriptional signatures, and reduced EMT and extracellular matrix remodelling signatures. These findings provide hypotheses for further mechanistic validation.

cancer biology↗