Cooperative Modular Representation Learning (CMRL), an uncertainty-gated multimodal framework that dynamically regulates inter-modality information flow based on sample-level epistemic uncertainty estimated via Evidential Deep Learning (EDL), is presented.
Abstract
Accurate prognosis in lung adenocarcinoma (LUAD) requires integration of high-dimensional transcriptomic profiles with compact but clinically stable patient covariates. Naïve fusion strategies allow the high-variance RNA-seq modality to dominate learned representations, suppressing clinical signal. We present Cooperative Modular Representation Learning (CMRL), an uncertainty-gated multimodal framework that dynamically regulates inter-modality information flow based on sample-level epistemic uncertainty estimated via Evidential Deep Learning (EDL). Each modality encoder produces a latent embedding and a scalar uncertainty score; an adaptive communication gate controls how much each module updates its representation from messages sent by the other module. A Variational Information Bottleneck (VIB) on the transcriptomic encoder further suppresses noise in the high-dimensional genomic latent space. CMRL is evaluated via 5-fold stratified cross-validation on 490 TCGA-LUAD patients with matched RNA-seq (504 features) and clinical data. It achieves a concordance index (C-index) of 0.732 ± 0.024, AUROC of 0.772 ± 0.019, and AUPRC of 0.773 ± 0.056 for 3-year survival prediction, outperforming a concatenation-fusion baseline (C-index 0.656), RNA-only (0.711), and clinical-only (0.670) variants, as well as several published LUAD survival models including CustOmics (0.625) and a whole-slide imaging method (0.675). An ablation study confirms that the uncertainty gate and evidential heads each contribute independently to the gain. Calibration analysis yields an Expected Calibration Error of 0.122, and uncertainty-stratified evaluation shows that low-uncertainty patients achieve AUROC 0.795 versus 0.681 for high-uncertainty patients, providing interpretable evidence that the gate mechanism is functioning as intended.
Accurate survival prediction is crucial for precision oncology, yet it faces challenges due to the neglect of clinical priors and high computational complexity. We propose TriBind-Mamba, a tri-modal framework integrating Clinical Knowledge Prompting (CKP) and selective State Space Models (SSMs). By transforming structured clinical records into semantic narratives using Large Language Models (LLMs), our model provides high-level context for morphological and molecular features. TriBind-Mamba efficiently processes gigapixel whole slide images and transcriptomic profiles with linear complexity, achieving state-ofthe-art performance (Overall C-index of 0.664) across five TCGA cohorts while significantly reducing computational overhead. Interpretability is enhanced by integrating human-readable clinical knowledge prompts, biologically meaningful pathway-level transcriptomic tokens, and WSI attention heatmaps that project model-derived importance scores back onto histopathological regions. These analyses suggest that TriBind-Mamba focuses on prognostically relevant malignant areas, providing a more transparent basis for multimodal survival prediction.
A multi-tier, explainable AI framework designed to risk-stratify patients and predict overall survival using clinical and genomic covariates is developed and demonstrates that explainable machine learning models can robustly predict survivability and highlight actionable features for oncology dashboards.
Lung cancer results in roughly 1.8 million fatalities annually worldwide, with non-small cell lung cancer (NSCLC) comprising the majority of cases. Despite advancements in treatment, survival stratification remains challenging due to intratumoral heterogeneity inadequately captured by conventional descriptors. Standard radiomic and deep learning techniques regard imaging features as independent quantities, overlooking structured interactions between tumor characteristics. We evaluate whether structured proxy features can enhance multimodal NSCLC survival prediction by augmenting pretreatment computed tomography (CT) representations, radiomics, and clinical variables with six simulation-derived features designed to capture interactions between heterogeneity and morphology.
A radiomic-parameterized cellular automaton generates growth-rate and necrosis-ratio proxy features from baseline CT by using entropy and sphericity to compute low-dimensional proxy parameters. The imaging backbone is a Transformer-based Masked Autoencoder (TMAE), which was chosen after a systematic evaluation with alternative encoders within the same pipeline and provides attention-based visualizations that highlight tumor regions receiving higher model attention.
On the public Lung1 cohort (
n
= 390), the primary four-modality fusion attained a C-index of 0.641 (iAUC 0.731, log-rank
p
< 0.001). The primary result compares favorably with prior multimodal results on Lung1 (C-index 0.631; iAUC 0.592 [15]) under a comparable evaluation protocol, while a separate exploratory coefficient-optimization analysis achieved a best observed C-index of 0.662 (iAUC 0.748).
These results indicate that, in addition to conventional radiomic, deep, and clinical representations within the Lung1 benchmark, simulation-derived proxy features may provide complementary predictive information within this fixed Lung1 benchmark. By integrating structured tumor-dynamics-inspired descriptors with modern volumetric CT representations, the framework provides a practical approach for retrospective relative risk ranking from routinely acquired pretreatment imaging and establishes a foundation for future repeated-split, external-cohort, and calibration studies.
Huu Phong Nguyen, Delower Hossain, Ehsan Saghapour et al.· Frontiers in Artificial Inte...· 0 citations
Predicting immunotherapy response from single-cell data remains difficult due to patient-level labels, extreme class imbalance, and highly heterogeneous macrophage states. We present a Multiple Instance Learning (MIL) framework that treats each patient as a bag of macrophage embeddings derived from a single-cell RNA foundation model. The architecture incorporates an attention-based pooling mechanism with reduced model complexity, dropout-enhanced regularization and explicit attention penalties to improve stability in small-sample regimes. To address imbalanced clinical datasets, MIL outputs are optimized with a combined focal loss and supervised contrastive objective that simultaneously sharpens class boundaries and improves representation clustering. Across three cancer datasets, this approach outperforms pseudobulk aggregation, embedding baselines and standard MIL variants. Attention-weighted attribution and transcriptional regulatory analysis reveal distinct macrophage programs, interferon and antigen-presentation networks in responders versus hypoxia-linked regulatory modules in non-responders. This shows the potential of MIL to uncover predictive and mechanistically interpretable immune states.
In biomarker discovery, access to sufficient quantities of condition-specific transcriptomic data is often limited by cohort size, privacy concerns, and domain shift between normal and condition populations. Generative modeling can augment scarce cohorts and probe distributional transitions. Furthermore, synthetic transcriptome generation can support differential expression analyses, machine learning, privacy-preserving data sharing, benchmarking, and hypothesis generation in translational bioinformatics workloads in fields such as oncology. Here, we present Hoike, a framework that combines a crossdomain Joint-Embedding Predictive Architecture (JEPA) with a latent diffusion model to generate condition-specific bulk transcriptomes from a normal reference context. In Hoike, normal tissue profiles provide continuous conditioning signals, while the model learns disease-linked shifts in latent space and reconstructs gene-level expression in log2(TPM+1) space. The implementation supports paired normal-condition training, tissuealigned conditioning, and constrained non-negative decoding for biologically valid outputs. We describe the architecture, objective design, and evaluation protocol used in this work across GTEx-derived normal references and multiple TCGA condition cohorts as a case study. This serves as the technical specification of the Hoike framework and its reproducible analysis workflow.
This work proposes MultiSigBERT, a unified framework for multimodal sequential survival modeling in oncology based on path signature representations that achieves a concordance index of 0.743 on an independent test set, demonstrating the benefit of jointly modeling multimodal temporal dynamics together with patient-level geometric structure for survival prediction.
Paul Minchella, Stéphane Chrétien, Guillaume Metzler et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
