Palo Alto Institute

To develop and validate a deep learning model for the detection of aberrant anterior tibial artery (AATA) on axial T2-weighted knee MRI, given the surgical relevance of unrecognized AATA and the lack of automated detection tools. This retrospective study included 70,260 MRI images from 2315 examinations (1441 without AATA and 874 with AATA) collected after institutional review board approval. Musculoskeletal radiologists performed image-level annotations. Data were split at the patient level into training, validation, and internal test sets; an independent dataset from another institution served as an external test set. A convolutional neural network was implemented in Python and PyTorch. Model performance was assessed at the patient level. At the slice level, the model achieved an F1-score of 0.838 on the internal test set. Patient-level classification using the validation-derived threshold (0.17) yielded F1-scores of 0.966 on the validation set, 0.979 on the internal test set, and 0.786 on the external test set. The area under the receiver operating characteristic curve for the external cohort was 0.97, indicating strong generalization despite a decrease in precision due to false positives. To our knowledge, this is the first study to apply artificial intelligence for automated detection of AATA on knee MRI. The proposed deep learning model performs this task with high sensitivity. Despite reduced precision in the external cohort, it demonstrates strong potential for enhancing preoperative risk assessment and surgical planning. Broader multicenter validation is warranted before clinical deployment.

Generalist biological artificial intelligence (GBAI) represents a transformative approach to modeling the 'language of life'-the flow of information from DNA to cellular function. This Review synthesizes rapid advances in biological AI to interpret and generate DNA, RNA, proteins and cellular systems. We chart a course toward comprehensive systems that can concurrently process and predict across these domains, performing several critical biological tasks simultaneously. Substantial opportunities lie in synergizing language and structural AI, leveraging specialized models and improving AI agents for autonomous discovery. After addressing challenges in data, biological complexity, scalability and experimental validation, GBAI has the potential to deepen our understanding of disease pathways and biomarkers, advance automated therapeutic design and evaluation, and integrate within virtual cells to meaningfully simulate biological activity.

We present a set of Grand Challenges for predictive modeling in small molecule drug discovery, with the goal of defining, prioritizing, and quantifying the areas where computation can have transformative impact. Rather than offering another broad survey of methods, this paper articulates specific scientific and technical problems that limit progress today and proposes measurable criteria by which advances can be judged. Our objective is to align researchers, investors, and industry leaders around the challenges that matter most for advancing real drug discovery programs. This work seeks to broaden participation in drug discovery by providing a clear roadmap for contributors from the growing areas of artificial intelligence, machine learning, robotics, high-performance computing, and quantum computing, where tools are rapidly advancing but are often disconnected from the practical realities of medicinal chemistry and pharmacology. The insights presented here draw on extensive discussions with experienced drug hunters, computational method developers, and thought leaders across biotech, pharma, software, and venture capital, as well as lessons learned from our own drug discovery efforts. While there is substantial enthusiasm (particularly around AI) for revolutionizing drug discovery, this moment demands sharper problem definition. Without clearly articulated challenges and performance metrics, computational innovation risks optimizing for benchmarks rather than for translational impact. We therefore identify Grand Challenges across four domains: Chemistry, Structure, Energy, and Pharmacology. For each domain, we present a common framework: a well-defined challenge, the underlying physical principles, its relevance to drug discovery decision-making, the current state of the field, and quantitative metrics that define progress. By grounding computational ambition in concrete scientific problems, we aim to catalyze sustained, measurable advances rather than episodic waves of enthusiasm.

The gastrointestinal tract harbours a vast chemical diversity of small molecules, consisting of dietary nutrients, microorganism-derived metabolites and metabolic products of the host. The latest evidence highlights a direct involvement of different metabolites in the diverse aetiologies of intestinal diseases, ranging from inflammatory to metabolic and neoplastic conditions. The accessibility of the gastrointestinal tract to oral intervention suggests that fine-tuning the levels of intestinal metabolites might be a promising and currently underutilized therapeutic strategy. Here, we provide a conceptual overview of the recurring mechanistic themes by which metabolites shape the biology of immune cells, epithelium and neurons of the gastrointestinal tract. Additionally, we classify metabolites according to possible categories of therapeutic intervention, and summarize the latest preclinical and clinical data unveiling the roles of intestinal metabolites in the pathophysiology of major diseases of the gastrointestinal tract, including inflammatory bowel disease, irritable bowel syndrome, colorectal cancer, enteric infection, food allergy, coeliac disease, as well as obesity and metabolic syndrome. In each case, we provide an overview of the mechanisms by which intestinal metabolites have been associated with disease aetiology. In addition, we discuss possible metabolite-based strategies for intervention. Our overall goal is to provide a roadmap towards developing metabotherapies for intestinal disease.

To describe a case of bilateral, multiple branch retinal artery occlusions (BRAO) in a woman on pembrolizumab immunotherapy for metastatic uterine cancer. Case report of one patient and literature review. A female in her mid-40s with obesity, hypertension, type 2 diabetes, and metastatic uterine cancer status post chemotherapy, debulking surgery, and whole-brain radiation presented with bilateral worsening of vision and a symptomatic scotoma in her right eye. Ten months prior to presentation, she was found to have recurrence of her cancer and was started on immunotherapy with pembrolizumab and lenvatinib. Dilated fundus examination showed vascular tortuosity, intraretinal hemorrhages, microaneurysms, and macular edema in both eyes. The macular edema was attributed to diabetic retinopathy and was treated with intravitreal bevacizumab. Although there was improvement of edema on optical coherence tomography, her vision did not significantly improve. Over sixteen months, her vision worsened from 20/40 and 20/50 to 20/200 and 20/150 in the right and left eye, respectively. Fluorescein angiography was performed and revealed multiple, bilateral branch retinal artery occlusions, symmetrically involving the superior arterial vessels in both eyes. Extensive imaging and laboratory work-up was negative. Her BRAOs were attributed to pembrolizumab. Pembrolizumab can be associated with multiple branch retinal artery occlusions. Fluorescein angiography is a useful tool in distinguishing between etiologies of retinal vascular disease. Oncologic and ophthalmologic providers should be cognizant of this rare cause of vision loss and screen appropriately.

The clinical success of cancer drug candidates depends on efficacy across many different individuals. Because xenografts are challenging to scale, we currently rely on a limited set of in vivo preclinical models. Here, to address this limitation, we introduce GENEVA, a scalable single-cell-resolution platform for measuring responses to drug perturbations. GENEVA models cancer genetic diversity by combining multiple patient-derived cell lines and cancer cell lines into pooled three-dimensional cultures and xenograft models, allowing us to study drug responses across a wide range of genetic backgrounds within a single experiment. We apply GENEVA to investigate KRAS-G12C inhibitors and demonstrate that mitochondrial activation is a key driver of cell death following KRAS inhibition, while epithelial-to-mesenchymal transition is a prominent resistance mechanism. These findings highlight the utility of GENEVA to identify therapeutic targets and optimize combination therapies with the potential to bridge the gap between preclinical cancer models and patient outcomes.

Global healthcare providers are exploring the use of large language models (LLMs) to provide medical advice to the public. LLMs now achieve nearly perfect scores on medical licensing exams, but this does not necessarily translate to accurate performance in real-world settings. We tested whether LLMs can assist members of the public in identifying underlying conditions and choosing a course of action (disposition) in ten medical scenarios in a controlled study with 1,298 participants. Participants were randomly assigned to receive assistance from an LLM (GPT-4o, Llama 3, Command R+) or a source of their choice (control). Tested alone, LLMs complete the scenarios accurately, correctly identifying conditions in 94.9% of cases and disposition in 56.3% on average. However, participants using the same LLMs identified relevant conditions in fewer than 34.5% of cases and disposition in fewer than 44.2%, both no better than the control group. We identify user interactions as a challenge to the deployment of LLMs for medical advice. Standard benchmarks for medical knowledge and simulated patient interactions do not predict the failures we find with human participants. Moving forward, we recommend systematic human user testing to evaluate interactive capabilities before public deployments in healthcare.

Aneuploidy, a hallmark of cancer, alters chromosome copy numbers and with that the abundance of hundreds of proteins. Evidence suggests that levels of proteins encoded on affected chromosomes are often buffered toward their abundances observed in diploids. Despite its prevalence, the molecular mechanisms driving this protein dosage compensation remain largely unknown. It is unclear whether all proteins are buffered similarly, what factors determine buffering, and whether dosage compensation varies across different cell lines or tumor types. Moreover, its potential adaptive advantage and therapeutic relevance remain unexplored. We established a novel approach to quantify protein dosage buffering in a gene copy number-dependent manner, showing that dosage compensation is widespread but variable in cancer samples. By developing multifactorial machine learning models, we identify gene dependency, protein complex participation, haploinsufficiency, and mRNA decay as key predictors of buffering. We show that dosage compensation affects oncogenic potential and that higher buffering correlates with reduced proteotoxic stress and increased drug resistance. These findings highlight protein dosage compensation as a crucial regulatory mechanism with therapeutic potential in aneuploid cancers.

Patients with moderate-to-severe atopic dermatitis (AD) who discontinue dupilumab need additional therapeutic options. Lebrikizumab's distinct mechanism of action may provide that alternative treatment. We evaluated efficacy and safety of lebrikizumab in adults and adolescents with moderate-to-severe AD previously treated with dupilumab. In the open-label ADapt trial, patients who discontinued dupilumab due to inadequate response, adverse events (AEs)/intolerance, or other reasons received lebrikizumab 250-mg every 2weeks (Q2W) following 500-mg loading dose at baseline/week2, through week16, with optional topical therapy. From weeks16-24, responders (≥ 75% improvement in Eczema Area and Severity Index [EASI75] or Investigator's Global Assessment score 0/1 with ≥ 2-point improvement from baseline) received lebrikizumab every 4weeks; inadequate responders continued lebrikizumab Q2W. The primary endpoint was EASI75 at week16 in the intent-to-treat population; EASI75 was also analyzed by reason for dupilumab discontinuation. Secondary and exploratory efficacy endpoints were assessed throughout. Among the 86 patients enrolled, primary reasons for stopping dupilumab were inadequate response (n = 48, 55.8%), AEs/intolerance (n = 14, 16.3%), and other reasons (n = 24, 27.9%). Fifty-nine patients (68.6%) completed week16; 52 patients (60.5%) completed week24. At weeks16 and 24, respectively, response rates were 57.4% (35/61) and 60.0% (33/55) for EASI75; 53.2% (25/47) and 61.5% (24/39) for Pruritus Numeric Rating Scale ≥ 4-point improvement; and 83.0% (44/53) and 83.0% (39/47) for Dermatology Life Quality Index ≥ 4-point improvement. Most treatment-emergent AEs were mild/moderate. Serious AEs and discontinuations due to AEs were reported by 2 (2.3%) and 5 (5.8%) patients, respectively. Of the 10 patients who discontinued dupilumab due to eye-related events, facial dermatitis, or inflammatory arthritis, none reported similar events with lebrikizumab. Results suggest that lebrikizumab provides meaningful improvements in skin clearance, itch, and quality of life in dupilumab-experienced patients with moderate-to-severe AD, with a safety profile consistent with other lebrikizumab phase3 trials. ClinicalTrials.gov identifier, NCT05369403.