Biology subjects

Foroughi pour, A.

Publications and source records attributed to Foroughi pour, A..

5 recordsLinked to original sources

bioRxiv2026

ExpertTwins: Auditable Long Horizon Scientific Panels through Grounded Expertise

Multi-agent language-model systems increasingly assign roles to agents and let them deliberate. A role label, however, does not specify which evidence an agent can bring or which standard it applies before accepting a conclusion. Agents that read one another's outputs tend to converge. We present ExpertTwins, a framework in which each seat of a scientific panel is defined by an inspectable evidence policy. Discipline seats retrieve under field-specific queries, and scientist-associated seats retrieve from attributed publications and a citation-derived scholarly neighborhood. Private initial packets, structured exchange, cited-source disclosure, deterministic quotation checks, and peer source review turn deliberation into an auditable record, which we also render as interactive idea-flow maps. Scientist associations denote publication-grounded perspectives, not replicas or endorsements of named researchers. In a nine-seat, four-turn neuroblastoma panel, 438 of 439 claims passed all attached quotation checks, and seats withdrew or defended positions when challenged on their evidence. A ten-turn panel sustained evidence acquisition and differentiated recommendations without conversational collapse. Across six base models given byte-identical initial packets, four sustained 97.9-99.8% quotation grounding, and at least 91% on every turn, whereas Haiku 4.5 failed by paraphrase and GPT-5 mini by silence. Agreement among strong models fell from 31% of claims at turn one to 5% at turn four. With follow-up questions held fixed, it stayed at 28-33% through turn three and fell to 20% only at the closing turn. Stress runs showed the collapse guardrail escalating through its full intervention ladder when consensus was forced, but missing deference to one unusually strong seat. On idea-flow maps, the forced consensus appeared as parallel lanes of identical ideas, although quotation grounding and participation balance stayed at their maxima. We also reimplemented two close published systems and ran them on matched models, seats, and questions. A multi-persona debate cited only real records, yet none of its cited sentences carried a checkable quotation and it had no way to abstain. An author-memory fusion system ran only after its fact checker was split by seat to fit in context. Its fused answer kept about two-thirds of the fact-checked specialist claims, answered 36 of 60 question parts on which our seats had abstained, and showed the reader 19% of its agents' disagreements. In a computational pathology case study, six scientist-associated agents built distinct pipelines to predict MYCN amplification from whole-slide images. Across six base models, seat-by-model interaction explained 57% of internal variance, and internal rank poorly predicted external rank (r = 0.18 across 36 nominees). These results establish differentiated, source-traceable scientific reasoning as an operational capability and identify the base model as a parameter that must be reported with the protocol.

bioinformatics↗

bioRxiv2025

Prediction of Outcome from Spatial Protein Profiling of Triple-Negative Breast Cancers

In tumors, reciprocal spatial interactions between immune cells, their mediators, the extracellular matrix, and mutated neoplastic cells impact all aspects of treatment resistance. The operational mechanisms of these interactions are foundational for developing insights and targets for cancer therapy and prevention. Spatial quantification of the tumor microenvironment system from image data has untapped potential for patient stratification. Here, we present SparTile, a powerful computational approach for analysis of multiplex proteomics images to reveal clinically relevant structural organization in the tumor microenvironment. SparTile enables the robust unbiased identification and characterization of tumor microenvironments based on spatial relationships among protein markers without the need for cell segmentation or classification. Applied to tissues of patients with triple negative breast cancer (TNBC), an aggressive subtype of breast cancer, SparTile identified repeatable microenvironments with specific cellular relationships. These included microenvironments characterized by risk markers such as Ki67+ (p-value= 0.052) and vimentin+ (p-value<0.01) tumor cells, which correlate with poor survival. Furthermore, myeloid markers in an MX1 positive tumor environment correlate with shorter survival (p-value=0.04). Finally, we identified a relative distance between tumor and myeloid cell markers as a strong prognostic risk factor in multivariate Cox models (p-value=0.01). This distance metric was externally validated on two datasets of breast cancer multiplex images (p-values<0.01). Thus, unbiased protein-based and segmentation-free spatial analysis is effective for identifying clinically relevant biomarkers from multiplex tumor images and identifies novel predictive biology.

cancer biology↗

bioRxiv2023

SAMPLER: Empirical distribution representations for rapid analysis of whole slide tissue images

Deep learning has revolutionized digital pathology, allowing for automatic analysis of hematoxylin and eosin (H&E) stained whole slide images (WSIs) for diverse tasks. In such analyses, WSIs are typically broken into smaller images called tiles, and a neural network backbone encodes each tile in a feature space. Many recent works have applied attention based deep learning models to aggregate tile-level features into a slide-level representation, which is then used for slide-level prediction tasks. However, training attention models is computationally intensive, necessitating hyperparameter optimization and specialized training procedures. Here, we propose SAMPLER, a fully statistical approach to generate efficient and informative WSI representations by encoding the empirical cumulative distribution functions (CDFs) of multiscale tile features. We demonstrate that SAMPLER-based classifiers are as accurate or better than state-of-the-art fully deep learning attention models for classification tasks including distinction of: subtypes of breast carcinoma (BRCA: AUC=0.911 {+/-} 0.029); subtypes of non-small cell lung carcinoma (NSCLC: AUC=0.940{+/-}0.018); and subtypes of renal cell carcinoma (RCC: AUC=0.987{+/-}0.006). A major advantage of the SAMPLER representation is that predictive models are >100X faster compared to attention models. Histopathological review confirms that SAMPLER-identified high attention tiles contain tumor morphological features specific to the tumor type, while low attention tiles contain fibrous stroma, blood, or tissue folding artifacts. We further apply SAMPLER concepts to improve the design of attention-based neural networks, yielding a context aware multi-head attention model with increased accuracy for subtype classification within BRCA and RCC (BRCA: AUC=0.921{+/-}0.027, and RCC: AUC=0.988{+/-}0.010). Finally, we provide theoretical results identifying sufficient conditions for which SAMPLER is optimal. SAMPLER is a fast and effective approach for analyzing WSIs, with greatly improved scalability over attention methods to benefit digital pathology analysis.

bioinformatics↗

bioRxiv2022

A pan-cancer PDX histology image repository with genomic and pathological annotations for deep learning analysis

Patient-derived xenografts (PDXs) model human intra-tumoral heterogeneity in the context of the intact tissue of immunocompromised mice. Histological imaging via hematoxylin and eosin (H&E) staining is performed on PDX samples for routine assessment and, in principle, captures the complex interplay between tumor and stromal cells. Deep learning (DL)-based analysis of large human H&E image repositories has extracted inter-cellular and morphological signals correlated with disease phenotype and therapeutic response. Here, we present an extensive, pan-cancer repository of nearly 1,000 PDX and paired human progenitor H&E images. These images, curated from the PDXNet consortium, are associated with genomic and transcriptomic data, clinical metadata, pathological assessment of cell composition, and, in several cases, detailed pathological annotation of tumor, stroma, and necrotic regions. We demonstrate that DL can be applied to these images to classify tumor regions and to predict xenograft-transplant lymphoproliferative disorder, the unintended outgrowth of human lymphocytes at the transplantation site. This repository enables PDX-specific, investigations of cancer biology through histopathological analysis and contributes important model system data that expand on existing human histology repositories. We expect the PDXNet Image Repository to be valuable for controlled digital pathology analysis, both for the evaluation of technical issues such as stain normalization and for development of novel computational methods based on spatial behaviors within cancer tissues.

bioinformatics↗

bioRxiv2022

Integrative deep learning analysis improves colon adenocarcinoma patient stratification at risk for mortality

Colorectal cancers are the fourth most commonly diagnosed cancer and the second leading cancer in number of deaths. Many clinical variables, pathological features, and genomic signatures are associated with patient risk, but reliable patient stratification in the clinic remains a challenging task. Here we assess how image, clinical, and genomic features can be combined to predict risk. We first observe that deep learning models based only on whole slide images (WSIs) from The Cancer Genome Atlas accurately separate high risk (OS<3years, N=38) from low risk (OS>5years, N=25) patients (AUC=0.81{+/-}0.08, 5year survival p-value=2.13e-25, 5year relative risk=5.09{+/-}0.05) though such models are less effective at predicting OS for moderate risk (3years<OS<5years, N=45) patients (5year survival p-value=0.5, 5year relative risk=1.32{+/-}0.09). However, we find that novel integrative models combining whole slide images, clinical variables, and mutation signatures can improve patient stratification for moderate risk patients (5year survival p-value=6.69e-30, 5year relative risk=5.32{+/-}0.07). Our integrative model combining image and clinical variables is also effective on an independent pathology dataset generated by our team (3year survival p-value=1.14e-09, 5year survival p-value=2.15e-05, 3year relative risk=3.25{+/-}0.06, 5year relative-risk=3.07{+/-}0.08). The integrative model substantially outperforms models using only images or only clinical variables, indicating beneficial cross-talk between the data types. Pathologist review of image-based heatmaps suggests that nuclear shape, nuclear size pleomorphism, intense cellularity, and abnormal structures are associated with high risk, while low risk regions tend to have more regular and small cells. The improved stratification of colorectal cancer patients from our computational methods can be beneficial for preemptive development of management and treatment plans for individual patients, as well as for informed enrollment of patients in clinical trials.

pathology↗