A Multimodal Generative AI Copilot for Human Pathology.

The field of computational pathology[1,2] has witnessed remarkable progress in the development of both task-specific predictive models and task-agnostic self-supervised vision encoders[3,4]. However, despite the explosive growth of generative artificial intelligence (AI), there has been limited study on building general purpose, multimodal AI assistants and copilots[5] tailored to pathology. Here we present PathChat, a vision-language generalist AI assistant for human pathology. We build PathChat by adapting a foundational vision encoder for pathology, combining it with a pretrained large language model and finetuning the whole system on over 456,000 diverse visual language instructions consisting of 999,202 question-answer turns. We compare PathChat against several multimodal vision language AI assistants and GPT4V, which powers the commercially available multimodal general purpose AI assistant ChatGPT-4[7]. PathChat achieved state-of-the-art performance on multiple-choice diagnostic questions from cases of diverse tissue origins and disease models. Furthermore, using open-ended questions and human expert evaluation, we found that overall PathChat produced more accurate and pathologist-preferable responses to diverse queries related to pathology. As an interactive and general vision-language AI Copilot that can flexibly handle both visual and natural language inputs, PathChat can potentially find impactful applications in pathology education, research, and human-in-the-loop clinical decision making.

Generation of Rhesus Macaque Embryos with Expanded CAG Trinucleotide Repeats in the Huntingtin Gene.

Huntington's disease (HD) arises from expanded CAG repeats in exon 1 of the Huntingtin (HTT) gene. The resultant misfolded HTT protein accumulates within neuronal cells, negatively impacting their function and survival. Ultimately, HTT accumulation results in cell death, causing the development of HD. A nonhuman primate (NHP) HD model would provide important insight into disease development and the generation of novel therapies due to their genetic and physiological similarity to humans. For this purpose, we tested CRISPR/Cas9 and a single-stranded DNA (ssDNA) containing expanded CAG repeats in introducing an expanded CAG repeat into the HTT gene in rhesus macaque embryos. Analyses were conducted on arrested embryos and trophectoderm (TE) cells biopsied from blastocysts to assess the insertion of the ssDNA into the HTT gene. Genotyping results demonstrated that 15% of the embryos carried an expanded CAG repeat. The integration of an expanded CAG repeat region was successfully identified in five blastocysts, which were cryopreserved for NHP HD animal production. Some off-target events were observed in biopsies from the cryopreserved blastocysts. NHP embryos were successfully produced, which will help to establish an NHP HD model and, ultimately, may serve as a vital tool for better understanding HD's pathology and developing novel treatments.

Enhancing Gluten Network Formation and Bread-Making Performance of Wheat Flour Using Wheat Bran Aqueous Extract.

Wheat bran possesses diverse nutritional and functional properties. In this study, wheat bran aqueous extract (WBE) was produced and thoroughly characterized as a functional ingredient and improver for bakery application. The WBE contained 50.3% total carbohydrate, 24.5% protein, 13.0% ash, 6.7% soluble fiber, 2.9% insoluble fiber, and 0.5% ß-glucan. Notably, adding 7.5% WBE significantly increased the bread-specific volume to 4.84 cm3/g, compared with the control of 4.18 cm3/g. Adding WBE also resulted in a remarkable improvement in dough properties. The WBE-enriched dough showed increased peak, setback, breakdown, and final viscosities, along with higher storage and loss modulus. Scanning electron microscopy analysis further revealed that the WBE promoted the aggregation of protein and starch within the dough. The extractable gliadin to glutenin ratio increased with 5 and 7.5% WBE additions, compared with the control and 2.5% WBE addition. WBE did not significantly alter the starch gelatinization temperature or dough extension properties. These findings demonstrate that the inclusion of WBE in wheat flour is a promising approach for producing high-quality bread that is enriched with dietary fiber and protein.

Demographic bias in misdiagnosis by computational pathology models.

Despite increasing numbers of regulatory approvals, deep learning-based computational pathology systems often overlook the impact of demographic factors on performance, potentially leading to biases. This concern is all the more important as computational pathology has leveraged large public datasets that underrepresent certain demographic groups. Using publicly available data from The Cancer Genome Atlas and the EBRAINS brain tumor atlas, as well as internal patient data, we show that whole-slide image classification models display marked performance disparities across different demographic groups when used to subtype breast and lung carcinomas and to predict IDH1 mutations in gliomas. For example, when using common modeling approaches, we observed performance gaps (in area under the receiver operating characteristic curve) between white and Black patients of 3.0% for breast cancer subtyping, 10.9% for lung cancer subtyping and 16.0% for IDH1 mutation prediction in gliomas. We found that richer feature representations obtained from self-supervised vision foundation models reduce performance variations between groups. These representations provide improvements upon weaker models even when those weaker models are combined with state-of-the-art bias mitigation strategies and modeling choices. Nevertheless, self-supervised vision foundation models do not fully eliminate these discrepancies, highlighting the continuing need for bias mitigation efforts in computational pathology. Finally, we demonstrate that our results extend to other demographic factors beyond patient race. Given these findings, we encourage regulatory and policy agencies to integrate demographic-stratified evaluation into their assessment guidelines.

Size-structure-property relationship of wood particles in aqueous and dry insulative foams.

Three size-fractionated samples of pine beetle-killed wood particles were used to prepare lightweight insulative foams. The foams were produced by foam-forming an aqueous slurry containing wood particles (125-1000 µm), a polymer binder, and surfactant, followed by oven drying. The effect of wood particle size on the aqueous foam stability, structure, and performance of insulative foams was investigated. While all aqueous foams were highly stable, aqueous foam stability increased with decreasing particle size. For dry foams, the cell size distribution was similar for all particle sizes as it was primarily controlled by the surfactant; differences occurred within the cell wall structure. A size-structure-property relationship was identified using x-ray micro-computed tomography where smaller particles produced lighter cell wall frameworks, leading to lower densities and decreased thermal conductivity and compressive strength. Larger particles produced denser cell wall frameworks that were more resistant to deformation, although all dry foams had sufficient mechanical properties for use as insulation panels. Thermal conductivity for all wood particle size-fractionated samples was <0.047 W m-1 K-1 making the foams similar to expanded polystyrene/polyurethane and supporting their use as thermal insulation in buildings.

A visual-language foundation model for computational pathology.

The accelerated adoption of digital pathology and advances in deep learning have enabled the development of robust models for various pathology tasks across a diverse array of diseases and patient cohorts. However, model training is often difficult due to label scarcity in the medical domain, and a model's usage is limited by the specific task and disease for which it is trained. Additionally, most models in histopathology leverage only image data, a stark contrast to how humans teach each other and reason about histopathologic entities. We introduce CONtrastive learning from Captions for Histopathology (CONCH), a visual-language foundation model developed using diverse sources of histopathology images, biomedical text and, notably, over 1.17 million image-caption pairs through task-agnostic pretraining. Evaluated on a suite of 14 diverse benchmarks, CONCH can be transferred to a wide range of downstream tasks involving histopathology images and/or text, achieving state-of-the-art performance on histology image classification, segmentation, captioning, and text-to-image and image-to-text retrieval. CONCH represents a substantial leap over concurrent visual-language pretrained systems for histopathology, with the potential to directly facilitate a wide array of machine learning-based workflows requiring minimal or no further supervised fine-tuning.

Towards a general-purpose foundation model for computational pathology.

Quantitative evaluation of tissue images is crucial for computational pathology (CPath) tasks, requiring the objective characterization of histopathological entities from whole-slide images (WSIs). The high resolution of WSIs and the variability of morphological features present significant challenges, complicating the large-scale annotation of data for high-performance applications. To address this challenge, current efforts have proposed the use of pretrained image encoders through transfer learning from natural image datasets or self-supervised learning on publicly available histopathology datasets, but have not been extensively developed and evaluated across diverse tissue types at scale. We introduce UNI, a general-purpose self-supervised model for pathology, pretrained using more than 100 million images from over 100,000 diagnostic H&E-stained WSIs (>77 TB of data) across 20 major tissue types. The model was evaluated on 34 representative CPath tasks of varying diagnostic difficulty. In addition to outperforming previous state-of-the-art models, we demonstrate new modeling capabilities in CPath such as resolution-agnostic tissue classification, slide classification using few-shot class prototypes, and disease subtyping generalization in classifying up to 108 cancer types in the OncoTree classification system. UNI advances unsupervised representation learning at scale in CPath in terms of both pretraining data and downstream evaluation, enabling data-efficient artificial intelligence models that can generalize and transfer to a wide range of diagnostically challenging tasks and clinical workflows in anatomic pathology.

Analytical validation of NeXT Personal®, an ultra-sensitive personalized circulating tumor DNA assay.

We describe the analytical validation of NeXT Personal®, an ultra-sensitive, tumor-informed circulating tumor DNA (ctDNA) assay for detecting residual disease, monitoring therapy response, and detecting recurrence in patients diagnosed with solid tumor cancers. NeXT Personal uses whole genome sequencing of tumor and matched normal samples combined with advanced analytics to accurately identify up to ~1,800 somatic variants specific to the patient's tumor. A personalized panel is created, targeting these variants and then used to sequence cell-free DNA extracted from patient plasma samples for ultra-sensitive detection of ctDNA. The NeXT Personal analytical validation is based on panels designed from tumor and matched normal samples from two cell lines, and from 123 patients across nine cancer types. Analytical measurements demonstrated a detection threshold of 1.67 parts per million (PPM) with a limit of detection at 95% (LOD95) of 3.45 PPM. NeXT Personal showed linearity over a range of 0.8 to 300,000 PPM (Pearson correlation coefficient = 0.9998). Precision varied from a coefficient of variation of 12.8% to 3.6% over a range of 25 to 25,000 PPM. The assay targets 99.9% specificity, with this validation study measuring 100% specificity and in silico methods giving us a confidence interval of 99.92 to 100%. In summary, this study demonstrates NeXT Personal as an ultra-sensitive, highly quantitative and robust ctDNA assay that can be used to detect residual disease, monitor treatment response, and detect recurrence in patients.

Sox9 regulates alternative splicing and pancreatic beta cell function.

Despite significant research, mechanisms underlying the failure of islet beta cells that result in type 2 diabetes (T2D) are still under investigation. Here, we report that Sox9, a transcriptional regulator of pancreas development, also functions in mature beta cells. Our results show that Sox9-depleted rodent beta cells have defective insulin secretion, and aging animals develop glucose intolerance, mimicking the progressive degeneration observed in T2D. Using genome editing in human stem cells, we show that beta cells lacking SOX9 have stunted first-phase insulin secretion. In human and rodent cells, loss of Sox9 disrupts alternative splicing and triggers accumulation of non-functional isoforms of genes with key roles in beta cell function. Sox9 depletion reduces expression of protein-coding splice variants of the serine-rich splicing factor arginine SRSF5, a major splicing enhancer that regulates alternative splicing. Our data highlight the role of SOX9 as a regulator of alternative splicing in mature beta cell function.

Search of inhibitors of aldose reductase for treatment of diabetic cataracts using machine learning.

Purpose: Patients with diabetes mellitus have an elevated chance of developing cataracts, a degenerative vision-impairing condition often needing surgery. The process of the reduction of glucose to sorbitol in the lens of the human eye that causes cataracts is managed by the Aldose Reductase Enzyme (AR), and it is been found that AR inhibitors may mitigate the onset of diabetic cataracts. There exists a large pool of natural and synthetic AR inhibitors that can prevent diabetic complications, and the development of a machine-learning (ML) prediction model may bring new AR inhibitors with better characteristics into clinical use. Methods: Using known AR inhibitors and their chemical-physical descriptors we created the ML model for prediction of new AR inhibitors. The predicted inhibitors were tested by computational docking to the binding site of AR. Results: Using cross-validation in order to find the most accurate ML model, we ended with final cross-validation accuracy of 90%. Computational docking testing of the predicted inhibitors gave a high level of correlation between the ML prediction score and binding free energy. Conclusions: Currently known AR inhibitors are not used yet for patients for several reasons. We think that new predicted AR inhibitors have the potential to possess more favorable characteristics to be successfully implemented after clinical testing. Exploring new inhibitors can improve patient well-being and lower surgical complications all while decreasing long-term medical expenses.

Effects of six pyrimidine analogs on the growth of Tetrahymena thermophila and their implications in pyrimidine metabolism.

Tetrahymena are ciliated protists that have been used to study the effects of toxic chemicals, including anticancer drugs. In this study, we tested the inhibitory effects of six pyrimidine analogs (5-fluorouracil, floxuridine, 5'-deoxy-5-fluorouridine, 5-fluorouridine, gemcitabine, and cytarabine) on wild-type CU428 and conditional mutant NP1 Tetrahymena thermophila at room temperature and the restrictive temperature (37°C) where NP1 does not form the oral apparatus. We found that phagocytosis was not required for pyrimidine analog entry and that all tested pyrimidine analogs inhibited growth except for cytarabine. IC50 values did not significantly differ between CU428 and NP1 for the same analog at either room temperature or 37°C. To investigate the mechanism of inhibition, we used two pyrimidine bases (uracil and thymine) and three nucleosides (uridine, thymidine, and 5-methyluridine) to determine whether the inhibitory effects from the pyrimidine analogs were reversible. We found that the inhibitory effects from 5-fluorouracil could be reversed by uracil and thymine, from floxuridine could be reversed by thymidine, and from 5'-deoxy-5-fluorouridine could be reversed by uracil. None of the tested nucleobases or nucleosides could reverse the inhibitory effects of gemcitabine or 5-fluorouridine. Our results suggest that the five pyrimidine analogs act on different sites to inhibit T. thermophila growth and that nucleobases and nucleosides are metabolized differently in Tetrahymena.

A General-Purpose Self-Supervised Model for Computational Pathology.

Tissue phenotyping is a fundamental computational pathology (CPath) task in learning objective characterizations of histopathologic biomarkers in anatomic pathology. However, whole-slide imaging (WSI) poses a complex computer vision problem in which the large-scale image resolutions of WSIs and the enormous diversity of morphological phenotypes preclude large-scale data annotation. Current efforts have proposed using pretrained image encoders with either transfer learning from natural image datasets or self-supervised pretraining on publicly-available histopathology datasets, but have not been extensively developed and evaluated across diverse tissue types at scale. We introduce UNI, a general-purpose self-supervised model for pathology, pretrained using over 100 million tissue patches from over 100,000 diagnostic haematoxylin and eosin-stained WSIs across 20 major tissue types, and evaluated on 33 representative CPath clinical tasks in CPath of varying diagnostic difficulties. In addition to outperforming previous state-of-the-art models, we demonstrate new modeling capabilities in CPath such as resolution-agnostic tissue classification, slide classification using few-shot class prototypes, and disease subtyping generalization in classifying up to 108 cancer types in the OncoTree code classification system. UNI advances unsupervised representation learning at scale in CPath in terms of both pretraining data and downstream evaluation, enabling data-efficient AI models that can generalize and transfer to a gamut of diagnostically-challenging tasks and clinical workflows in anatomic pathology.

Analytic validation of NeXT Dx™, a comprehensive genomic profiling assay.

We describe the analytic validation of NeXT Dx, a comprehensive genomic profiling assay to aid therapy and clinical trial selection for patients diagnosed with solid tumor cancers. Proprietary methods were utilized to perform whole exome and whole transcriptome sequencing for detection of single nucleotide variants (SNVs), insertions/deletions (indels), copy number alterations (CNAs), and gene fusions, and determination of tumor mutation burden and microsatellite instability. Variant calling is enhanced by sequencing a patient-specific normal sample from, for example, a blood specimen. This provides highly accurate somatic variant calls as well as the incidental reporting of pathogenic and likely pathogenic germline alterations. Fusion detection via RNA sequencing provides more extensive and accurate fusion calling compared to DNA-based tests. NeXT Dx features the proprietary Accuracy and Content Enhanced technology, developed to optimize sequencing and provide more uniform coverage across the exome. The exome was validated at a median sequencing depth of >500x. While variants from 401 cancer-associated genes are currently reported from the assay, the exome/transcriptome assay is broadly validated to enable reporting of additional variants as they become clinically relevant. NeXT Dx demonstrated analytic sensitivities as follows: SNVs (99.4%), indels (98.2%), CNAs (98.0%), and fusions (95.8%). The overall analytic specificity was >99.0%.

Leser-Trélat Sign as a Marker for Underlying Pancreatic Cancer.

CASE PRESENTATION: Early diagnosis and rapid treatment of cancer is essential for good clinical outcomes for patients. In this case, an 85-year-old man presented with failure to thrive and was noted to have rapid-onset, multiple seborrheic keratoses (Leser-Trélat sign) on his chest and back. He was ultimately diagnosed with pancreatic cancer using computed tomography. DISCUSSION: Leser-Trélat sign is a rare cutaneous marker for underlying malignancy. Identification of this sign can help guide diagnostic imaging and lab work to identify an occult internal malignancy, resulting in more rapid diagnosis, earlier treatment, and potentially better clinical outcomes.

Algorithmic fairness in artificial intelligence for medicine and healthcare.

In healthcare, the development and deployment of insufficiently fair systems of artificial intelligence (AI) can undermine the delivery of equitable care. Assessments of AI models stratified across subpopulations have revealed inequalities in how patients are diagnosed, treated and billed. In this Perspective, we outline fairness in machine learning through the lens of healthcare, and discuss how algorithmic biases (in data acquisition, genetic variation and intra-observer labelling variability, in particular) arise in clinical workflows and the resulting healthcare disparities. We also review emerging technology for mitigating biases via disentanglement, federated learning and model explainability, and their role in the development of AI-based software as a medical device.