A foundation model for clinical-grade computational pathology and rare cancers detection.

The analysis of histopathology images with artificial intelligence aims to enable clinical decision support systems and precision medicine. The success of such applications depends on the ability to model the diverse patterns observed in pathology images. To this end, we present Virchow, the largest foundation model for computational pathology to date. In addition to the evaluation of biomarker prediction and cell identification, we demonstrate that a large foundation model enables pan-cancer detection, achieving 0.95 specimen-level area under the (receiver operating characteristic) curve across nine common and seven rare cancers. Furthermore, we show that with less training data, the pan-cancer detector built on Virchow can achieve similar performance to tissue-specific clinical-grade models in production and outperform them on some rare variants of cancer. Virchow's performance gains highlight the value of a foundation model and open possibilities for many high-impact applications with limited amounts of labeled training data.

Identifying patterns in amyotrophic lateral sclerosis progression from sparse longitudinal data.

The clinical presentation of amyotrophic lateral sclerosis (ALS), a fatal neurodegenerative disease, varies widely across patients, making it challenging to determine if potential therapeutics slow progression. We sought to determine whether there were common patterns of disease progression that could aid in the design and analysis of clinical trials. We developed an approach based on a mixture of Gaussian processes to identify clusters of patients sharing similar disease progression patterns, modeling their average trajectories and the variability in each cluster. We show that ALS progression is frequently nonlinear, with periods of stable disease preceded or followed by rapid decline. We also show that our approach can be extended to Alzheimer's and Parkinson's diseases. Our results advance the characterization of disease progression of ALS and provide a flexible modeling approach that can be applied to other progressive diseases.

Discovery of Parkinson's disease states and disease progression modelling: a longitudinal data study using machine learning.

BACKGROUND: Parkinson's disease is heterogeneous in symptom presentation and progression. Increased understanding of both aspects can enable better patient management and improve clinical trial design. Previous approaches to modelling Parkinson's disease progression assumed static progression trajectories within subgroups and have not adequately accounted for complex medication effects. Our objective was to develop a statistical progression model of Parkinson's disease that accounts for intra-individual and inter-individual variability and medication effects. METHODS: In this longitudinal data study, data were collected for up to 7-years on 423 patients with early Parkinson's disease and 196 healthy controls from the Parkinson's Progression Markers Initiative (PPMI) longitudinal observational study. A contrastive latent variable model was applied followed by a novel personalised input-output hidden Markov model to define disease states. Clinical significance of the states was assessed using statistical tests on seven key motor or cognitive outcomes (mild cognitive impairment, dementia, dyskinesia, presence of motor fluctuations, functional impairment from motor fluctuations, Hoehn and Yahr score, and death) not used in the learning phase. The results were validated in an independent sample of 610 patients with Parkinson's disease from the National Institute of Neurological Disorders and Stroke Parkinson's Disease Biomarker Program (PDBP). FINDINGS: PPMI data were download July 25, 2018, medication information was downloaded on Sept 24, 2018, and PDBP data were downloaded between June 15 and June 24, 2020. The model discovered eight disease states, which are primarily differentiated by functional impairment, tremor, bradykinesia, and neuropsychiatric measures. State 8, the terminal state, had the highest prevalence of key clinical outcomes including 18 (95%) of 19 recorded instances of dementia. At study outset 4 (1%) of 333 patients were in state 8 and 138 (41%) of 333 patients reached stage 8 by year 5. However, the ranking of the starting state did not match the ranking of reaching state 8 within 5 years. Overall, patients starting in state 5 had the shortest time to terminal state (median 2·75 [95% CI 1·75-4·25] years). INTERPRETATION: We developed a statistical progression model of early Parkinson's disease that accounts for intra-individual and inter-individual variability and medication effects. Our predictive model discovered non-sequential, overlapping disease progression trajectories, supporting the use of non-deterministic disease progression models, and suggesting static subtype assignment might be ineffective at capturing the full spectrum of Parkinson's disease progression. FUNDING: Michael J Fox Foundation.

DPVis: Visual Analytics With Hidden Markov Models for Disease Progression Pathways.

Clinical researchers use disease progression models to understand patient status and characterize progression patterns from longitudinal health records. One approach for disease progression modeling is to describe patient status using a small number of states that represent distinctive distributions over a set of observed measures. Hidden Markov models (HMMs) and its variants are a class of models that both discover these states and make inferences of health states for patients. Despite the advantages of using the algorithms for discovering interesting patterns, it still remains challenging for medical experts to interpret model outputs, understand complex modeling parameters, and clinically make sense of the patterns. To tackle these problems, we conducted a design study with clinical scientists, statisticians, and visualization experts, with the goal to investigate disease progression pathways of chronic diseases, namely type 1 diabetes (T1D), Huntington's disease, Parkinson's disease, and chronic obstructive pulmonary disease (COPD). As a result, we introduce DPVis which seamlessly integrates model parameters and outcomes of HMMs into interpretable and interactive visualizations. In this article, we demonstrate that DPVis is successful in evaluating disease progression models, visually summarizing disease states, interactively exploring disease progression patterns, and building, analyzing, and comparing clinically relevant patient subgroups.

Closed-loop optimization of fast-charging protocols for batteries with machine learning.

Simultaneously optimizing many design parameters in time-consuming experiments causes bottlenecks in a broad range of scientific and engineering disciplines1,2. One such example is process and control optimization for lithium-ion batteries during materials selection, cell manufacturing and operation. A typical objective is to maximize battery lifetime; however, conducting even a single experiment to evaluate lifetime can take months to years3-5. Furthermore, both large parameter spaces and high sampling variability3,6,7 necessitate a large number of experiments. Hence, the key challenge is to reduce both the number and the duration of the experiments required. Here we develop and demonstrate a machine learning methodology  to efficiently optimize a parameter space specifying the current and voltage profiles of six-step, ten-minute fast-charging protocols for maximizing battery cycle life, which can alleviate range anxiety for electric-vehicle users8,9. We combine two key elements to reduce the optimization cost: an early-prediction model5, which reduces the time per experiment by predicting the final cycle life using data from the first few cycles, and a Bayesian optimization algorithm10,11, which reduces the number of experiments by balancing exploration and exploitation to efficiently probe the parameter space of charging protocols. Using this methodology, we rapidly identify high-cycle-life charging protocols among 224 candidates in 16 days (compared with over 500 days using exhaustive search without early prediction), and subsequently validate the accuracy and efficiency of our optimization approach. Our closed-loop methodology automatically incorporates feedback from past experiments to inform future decisions and can be generalized to other applications in battery design and, more broadly, other scientific domains that involve time-intensive experiments and multi-dimensional design spaces.

Identifying unreliable predictions in clinical risk models.

The ability to identify patients who are likely to have an adverse outcome is an essential component of good clinical care. Therefore, predictive risk stratification models play an important role in clinical decision making. Determining whether a given predictive model is suitable for clinical use usually involves evaluating the model's performance on large patient datasets using standard statistical measures of success (e.g., accuracy, discriminatory ability). However, as these metrics correspond to averages over patients who have a range of different characteristics, it is difficult to discern whether an individual prediction on a given patient should be trusted using these measures alone. In this paper, we introduce a new method for identifying patient subgroups where a predictive model is expected to be poor, thereby highlighting when a given prediction is misleading and should not be trusted. The resulting "unreliability score" can be computed for any clinical risk model and is suitable in the setting of large class imbalance, a situation often encountered in healthcare settings. Using data from more than 40,000 patients in the Global Registry of Acute Coronary Events (GRACE), we demonstrate that patients with high unreliability scores form a subgroup in which the predictive model has both decreased accuracy and decreased discriminatory ability.

Vasopressin Administration Is Associated With Rising Serum Lactate Levels in Patients With Sepsis.

BACKGROUND: Vasopressin is used in conjunction with norepinephrine during treatment of patients with septic shock. Serum lactate is often used in monitoring of patients with sepsis; however, its importance as a therapeutic target is unclear. The objective of this study is to examine the relationship of vasopressin use on serum lactate levels in patients with sepsis. METHODS: This study uses electronic heath records available via the Medical Information Mart for Intensive Care III. Patients were required to have a serum lactate monitoring during the intensive care unit (ICU) stay. The treatment was the administration of vasopressin between hours 3 and 18 of the ICU stay. Analysis was performed using a matched design. RESULTS: Patients receiving vasopressin were more likely to have their serum lactate levels rise when compared to matched patients who did not receive vasopressin (odds ratio: 6.6; 95% confidence interval: 3.0-14.6, P < .001). Patients who received vasopressin had a median increase in serum lactate of 0.3 mmol/L, while patients who did not receive vasopressin had a median decrease in serum lactate of 0.7 mmol/L (P < .001). There was no statistically significant difference between the control and treated groups' lactate trajectories prior to possible administration of vasopressin (P = .15). The results did not change significantly when norepinephrine initiation was used as the index time. CONCLUSIONS: In patients with sepsis, the administration of vasopressin was associated with a statistically significant difference in lactate change over the course of 24 hours when compared to matched patients who did not receive vasopressin.

Opportunities and challenges of real-time release testing in biopharmaceutical manufacturing.

Real-time release testing (RTRT) is defined as "the ability to evaluate and ensure the quality of in-process and/or final drug product based on process data, which typically includes a valid combination of measured material attributes and process controls" (ICH Q8[R2]). This article discusses sensors (process analytical technology, PAT) and control strategies that enable RTRT for the spectrum of critical quality attributes (CQAs) in biopharmaceutical manufacturing. Case studies from the small-molecule and biologic pharmaceutical industry are described to demonstrate how RTRT can be facilitated by integrated manufacturing and multivariable control strategies to ensure the quality of products. RTRT can enable increased assurance of product safety, efficacy, and quality-with improved productivity including faster release and potentially decreased costs-all of which improve the value to patients. To implement a complete RTRT solution, biologic drug manufacturers need to consider the special attributes of their industry, particularly sterility and the measurement of viral and microbial contamination. Continued advances in on-line and in-line sensor technologies are key for the biopharmaceutical manufacturing industry to achieve the potential of RTRT. Related article: http://onlinelibrary.wiley.com/doi/10.1002/bit.26378/full.

A method for learning a sparse classifier in the presence of missing data for high-dimensional biological datasets.

MOTIVATION: This work addresses two common issues in building classification models for biological or medical studies: learning a sparse model, where only a subset of a large number of possible predictors is used, and training in the presence of missing data. This work focuses on supervised generative binary classification models, specifically linear discriminant analysis (LDA). The parameters are determined using an expectation maximization algorithm to both address missing data and introduce priors to promote sparsity. The proposed algorithm, expectation-maximization sparse discriminant analysis (EM-SDA), produces a sparse LDA model for datasets with and without missing data. RESULTS: EM-SDA is tested via simulations and case studies. In the simulations, EM-SDA is compared with nearest shrunken centroids (NSCs) and sparse discriminant analysis (SDA) with k-nearest neighbors for imputation for varying mechanism and amount of missing data. In three case studies using published biomedical data, the results are compared with NSC and SDA models with four different types of imputation, all of which are common approaches in the field. EM-SDA is more accurate and sparse than competing methods both with and without missing data in most of the experiments. Furthermore, the EM-SDA results are mostly consistent between the missing and full cases. Biological relevance of the resulting models, as quantified via a literature search, is also presented. AVAILABILITY AND IMPLEMENTATION: A Matlab implementation published under GNU GPL v.3 license is available at http://web.mit.edu/braatzgroup/links.html . CONTACT: braatz@mit.edu. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.