Citations

Automated diagnostic report generation lies at the core of clinical diagnosis and can alleviate clinician shortages. Existing diagnostic report generation methods have two major limitations: they rely on unimodal inputs (e.g., images), ignoring textual biomarkers like medical history, and lack proactive dialogue capabilities to elicit personalized clinical information. To address those issues, we propose a ProActive Multimodal Agentic (PAMA) system, which performs comprehensive disease analysis by examining biomarkers in diverse sources, including multi-view medical images, medical histories, and diagnostic conversations. Built upon a knowledge graph and recommendation-based dialogue architecture, PAMA actively initiates adaptive, multi-turn conversations with patients, which is integrated with visual data for robust and reliable report generation. Specifically, PAMA actively generates adaptive multi-turn questions to collect clinically relevant background information, and then fuses the resulting dialogue context with visual representations for robust diagnostic report generation. We validate our approach on two real-world benchmark datasets, MIMIC-CXR and IU-Xray, through extensive quantitative evaluations and comparisons with state-of-the-art baselines. Furthermore, we conduct user studies to assess the realism and clinical quality of the generated reports. Finally, we present real-world case studies to examine the performance of our system across diverse scenarios, demonstrating its robustness on scalability, complexity, and data variability.

Abstract Background Diagnosis and risk stratification in rare hematologic malignancies such as myeloproliferative neoplasms (MPNs) – polycythemia vera (PV), essential thrombocythemia (ET), and myelofibrosis (MF) – require expert review of longitudinal, heterogeneous clinical records. This process is cognitively demanding, inconsistently applied, and difficult to scale beyond tertiary centers. No automated phenotyping workflow currently exists for hematologic malignancies. Methods A HIPAA-compliant large language model (LLM) framework for phenotyping MPN was developed to integrate (i) rule-based retrieval of bone marrow biopsy reports, clinical notes, and structured laboratory results from the electronic health record (EHR); (ii) zero-shot extraction of diagnostic and prognostic variables from unstructured text using GPT-4 Turbo; (iii) a clinician-informed source-prioritization algorithm to reconcile conflicting multi-source data; (iv) WHO/ICC-criteria–based diagnostic classification; and (v) NCCN-based risk stratification using the conventional risk model for PV, IPSET-thrombosis for ET, and DIPSS, DIPSS-plus, and MIPSS70/MIPSS70+ v2 for MF. Patients were identified via MPN-related ICD-9/10 codes; cases met 2017 WHO criteria or had a hematologist-documented diagnosis, and controls did not. The cohort was split into a prompt-development set (n = 60) and a held-out test set (n = 450; 75 cases and 75 controls per disease). Ground truth was established by independent dual-clinician chart review with consensus adjudication. LLM performance was evaluated against the ground truth: variable-level extraction using accuracy, F1 score, and Cohen’s kappa; patient-level diagnostic classification using sensitivity, specificity, and Cohen’s kappa; and prognostic risk stratification (among confirmed cases) using accuracy, weighted F1 score, and quadratic-weighted Cohen’s kappa. Wilson 95% confidence intervals (CIs) were used for proportions and bootstrap 95% CIs with 500 resamples for F1 scores. Results The held-out test set included 450 patients (PV: 150; ET: 150; MF: 150) with pathology reports and structured laboratory results, and 172 patients (PV: 52; ET: 55; MF: 65) with clinical notes. From pathology reports, overall variable extraction accuracy and F1 score were 99% (95% CI, 98–100) and 1.00 (0.99–1.00) for PV, 100% (99–100) and 0.99 (0.96–1.00) for ET, and 100% (99–100) and 0.99 (0.97–1.00) for MF. From clinical notes, overall accuracy and F1 score were 96% (91–100) and 0.94 (0.85–1.00) for PV, 100% (100–100) and 1.00 (1.00–1.00) for ET, and 100% (99–100) and 0.98 (0.95–1.00) for MF. Diagnostic sensitivity was 100% (95% CI, 95.1–100.0) for PV, ET, and MF; specificity was 98.7% (92.8–99.8) for PV and 100% (95.1–100.0) for both ET and MF, with Cohen’s kappa of 0.99 for PV and 1.00 for ET and MF. Risk stratification accuracy was 100% with weighted F1 score of 1.00 and quadratic-weighted Cohen’s kappa of 1.00 across all three diseases. A pre-specified source-ablation analysis showed that pathology reports alone were sufficient for diagnosis (sensitivity 98.7% for PV, 100% for ET, 96.0% for MF; specificity 100% across all three subtypes) but inadequate for prognostication (accuracy 69.3% for PV, 93.3% for ET, 77.3% for MF). Adding clinical notes to pathology reports recovered full prognostic accuracy of 100% across all three diseases. Conclusions This first-in-class automated framework achieved expert-level performance for MPN diagnosis and risk stratification from real-world EHR data, establishing a foundation for scalable, standardized phenotyping in rare hematologic malignancies. Prospective, multi-site validation is warranted before clinical deployment.

Abstract Intercellular communication between multiple myeloma (MM) cells and osteoclast precursor cells (pre-OCs) contributes extensively to the occurrence and development of myeloma-related bone destruction. However, key interacting modes and the exchanged substances involved in this communication remain unclear. In this study, we discover that tunneling nanotubes (TNTs) directly connect MM and pre-OCs. Using the stable isotope labeling with amino acids in cell culture assay and a positive-negative double selection strategy, we identify filamin A (FLNA) as a major protein transported from MM to pre-OCs. FLNA acts as a molecular clutch linking extracellular matrix–bound MAC1 to the cytoskeleton, activating Rho and MAPK signaling pathways and promoting F-actin polymerization, which subsequently enhances osteoclast differentiation by modulating cellular stiffness, traction force, and deformability. In addition, FLNA directly binds vinculin and promotes its recruitment to podosomes, thereby enhancing the functions of podosomes and the bone resorption capacity of osteoclasts. The conditional depletion of Flna in mice suppresses podosome activity and reduces stiffness, traction force, and deformability in pre-OCs, leading to significantly impaired osteoclast differentiation and increased bone mass. In the Vk∗MYC mouse model of myeloma, the administration of the TNT inhibitor latrunculin B disrupts FLNA transport to osteoclasts and alleviates osteolytic bone disease. These findings highlight the critical role of MM-transferred FLNA in osteoclastogenesis and suggest that targeting TNTs may represent a therapeutic strategy to limit pathological bone resorption associated with MM.

Abstract AI-powered pathology foundation models provide general-purpose representations of histopathological images by encoding image tiles into numerical embeddings. However, these embeddings are not directly interpretable in biological or clinical terms and must be translated into biologically meaningful features, such as cell-type composition or gene expression, to enable downstream clinical applications. To bridge this gap, we developed STpath, a framework that integrates histopathology image embeddings derived from existing pathology foundation models with matched, spatially resolved transcriptomics data. STpath consists of cancer-specific XGBoost models trained to infer cell-type compositions and gene expression from histopathology image tiles. We evaluated STpath in colorectal and breast cancer datasets and showed that it provides accurate estimates of the composition of major cell types and the expression of a subset of genes, with further performance gains achieved by combining embeddings from multiple foundation models. Finally, we demonstrated that STpath inferred features that can be used in downstream studies to evaluate their associations with clinical outcomes.

Abstract Recent updates to monocyte count thresholds recognize oligomonocytic chronic myelomonocytic leukemia (OM-CMML) as an early form of CMML. However, the clinical validity of these changes remains uncertain without incorporating biological and genomic factors. In this study, we analyzed a cohort of 911 patients (249 with OM-CMML, 359 with overt CMML, and 303 with myelodysplastic syndromes) using unsupervised clustering to evaluate the role of genomic determinants in refining CMML diagnosis. Our findings show that CMML molecular signatures (biallelic TET2 mutations or SRSF2-TET2 comutations) are linked to a distinct transcriptome, monocytic bias, classical monocytosis, and a higher risk of progression to overt CMML in OM-CMML cases. We developed a weighted genomic model and diagnostic workflow showing that combining genomic signatures with bone marrow monocyte frequencies in OM-CMML more accurately predicts progression to overt CMML. These findings support integrating genomic determinants and our clinic-ready diagnostic workflow into the CMML diagnostic framework to improve accuracy. Significance: Through comprehensive clinical and genomic profiling of a large patient cohort, alongside immunophenotypic and transcriptional cellular analyses, this study provides evidence that incorporating genomic determinants into the diagnostic criteria for OM-CMML improves diagnostic accuracy and refines the identification of early-stage CMML, thereby preventing misclassification.

Abstract Background: While mutations in the fms-like tyrosine kinase 3 (FLT3) gene are well-established adverse drivers in acute myeloid leukemia (AML), their role in myelodysplastic syndromes (MDS) and chronic myelomonocytic leukemia (CMML) remains insufficiently defined. Methods: We conducted a retrospective multicenter study of 115 patients with FLT3-mutated (FLT3mut) MDS (n=63) or CMML (n=52), comparing their clinical, genomic, and therapeutic characteristics at diagnosis or first referral to 677 FLT3 wild-type (FLT3wt) controls (MDS n=388; CMML n=289) using multivariate analyses and propensity score matching (PSM). Results: FLT3mut MDS and CMML exhibited a proliferative phenotype with higher leukocyte and neutrophil counts and lower platelet levels (p<0.02); in MDS, median white blood cell counts remained below 13,000/µL. In MDS, FLT3mut cases were enriched in the MDS with increased blasts subtype (57% vs 39%, p=0.01) and favorable-risk cytogenetics (64% vs 41%, p<0.01). Genomically, FLT3mut MDS showed higher frequencies of NRAS, KRAS, PTPN11, RUNX1 and CBL co-mutations (p<0.05), and a markedly lower prevalence of TP53 mutations (2% vs 36%, p<0.001). In CMML, FLT3mut cases were more often transfusion-dependent (50% vs 23%, p<0.01), classified as CMML-2 (39% vs 25%, p=0.04), and enriched in ASXL1 (68% vs 47%; p<0.01) and SETBP1 (24% vs 8%; p<0.01) mutations. Notably, they exhibited multi-hit TET2 mutations less frequently than FLT3wt CMML (27% vs 53%, p=0.01). After adjusting for age, sex, and allogeneic stem cell transplantation, FLT3 mutations were independently associated with worse outcomes. To further assess the prognostic impact of FLT3 mutations, we conducted PSM analyses using a 1:1 ratio of FLT3mut to FLT3wt patients, adjusting for key clinical and biological confounders. In MDS, matching was based on age, sex, bone marrow blast percentage, and cytogenetic risk category; in CMML, variables included age, CMML-2 subtype, peripheral blood counts, normal karyotype, and ASXL1 mutation status. After matching, FLT3mut MDS cases showed significantly shorter overall survival (OS; 19.9 vs 44.2 months, p=0.046) and leukemia-free survival (LFS; 13.8 vs 37.1 months, p=0.023) compared to FLT3wt patients. Similarly, in CMML, FLT3mut patients experienced inferior OS (30.4 vs 67.2 months, p=0.017) and LFS (17.2 vs 65.6 months, p=0.014). Competing risk models confirmed a markedly increased cumulative incidence of AML transformation in both FLT3mut MDS and CMML (p<0.001), without significant differences in the risk of non-leukemic death. Among FLT3mut cases, tyrosine kinase domain mutations (FLT3-TKD) were associated with superior survival compared to internal tandem duplications (FLT3-ITD; MDS hazard ratio [HR]: 0.28, p=0.003; CMML HR: 0.34, p=0.012). FLT3 allelic ratio (AR), assessed by conventional PCR in 63 patients (MDS n=34, CMML n=29), provided additional prognostic value. An AR ≥0.21 identified a subgroup with significantly shorter OS (MDS: 24.4 vs 38.8; CMML: 6.7 vs 30.4 months) and LFS (MDS: 9.0 vs 18.5 months; CMML: 5.8 vs 30.4 months; all p<0.05), supporting the utility of quantitative FLT3 assessment in clinical risk stratification. FLT3 inhibitors (FLT3i) were administered in combination with hypomethylating agents (HMA) to 16 (27%) FLT3mut MDS and 15 (33%) CMML patients. Compared to HMA, FLT3i treatment was associated with higher overall response rates (MDS: 69% vs 22%, p<0.01; CMML: 73% vs 35%, p=0.06). Molecular clearance, defined as post-treatment reduction in AR to ≤0.01, was observed more frequently with FLT3i than with HMA (MDS: 56% vs 10%, p=0.03; CMML: 60% vs 20%, p=0.09). FLT3 clearance was associated with prolonged survival. In MDS, median OS and LFS were 66.0 vs 14.5 (p=0.12) and 29.0 vs 9.9 months (p=0.056), respectively; in CMML, it was 37.6 vs 14.6 months for both OS (p=0.043) and LFS (p=0.032). Conclusions: This study expands the current understanding of FLT3mut MDS and CMML by characterizing their clinical, molecular, and therapeutic features in the largest multicenter cohort reported to date. These cases exhibit a proliferative phenotype, frequent co-mutations in RAS pathway genes, and an elevated risk of leukemic transformation. Quantitative assessment of FLT3 allelic burden refines prognostication beyond mutation status alone. The observed associations between FLT3 inhibitor use, molecular clearance, and improved outcomes support further evaluation of targeted strategies in selected patients.

Abstract RAS pathway (RASp) mutations induce proliferative features, and promote transformation in chronic myelomonocytic leukemia (CMML). However, the unique clonal landscape and hierarchy of distinct RASp mutations remain unexplored. To characterize the landscape, architecture, and implications of unique RASp mutations in CMML, we evaluated a cohort of 814 patients with CMML. We identified 461 RASp mutations among 342 patients (42%). N/KRAS and CBL mutations were the most common, frequently involved the P-loop or RING domains, respectively, and frequently appeared as dominant events (63% and 65%, respectively). BRAF, NF1, and PTPN11 mutations spanned throughout the gene structure, and frequently appeared as subclonal events (75%, 64%, and 59%, respectively). CBL mutations frequently occurred in codominance with SRSF2 and multihit TET2, and were enriched for KIT mutations. PTPN11 mutations more frequently co-occurred with SETBP1 and DNMT3A mutations, and were infrequently codominant with TET2 or ASXL1. RASp mutations predicted for shorter overall survival (hazard ratio [HR], 1.55; 95% confidence interval [CI], 1.15-2.07; P = .0075) and leukemia-free survival (LFS; HR, 1.67; 95% CI, 1.26-2.20; P = .0011), influenced outcomes of myelodysplastic and TET2 mutant CMML, and cooperated with IDH2 and RUNX1 mutations to induce shorter LFS. These data set the bases for refined genomic classifications of CMML, and underscore the need to develop RAS-directed therapies for patients with CMML.

Abstract Aggregation and nuclear depletion of the RNA binding protein TDP-43 are the crucial pathological features of amyotrophic lateral sclerosis (ALS) and inclusion body myositis (IBM), two degenerative diseases of the CNS and muscle. The loss of TDP-43 nuclear function results in the aberrant inclusion of cryptic exons in mRNA transcripts, leading to the expression of de novo proteins. Clonally expanded and highly differentiated CD8 + T cells have been observed in individuals with TDP-43 proteinopathies and therapeutics modulating the T cell response have recently been found to extend survival. However, the target antigens mediating T cell activation have remained elusive. Here, we investigate whether the de novo proteins induced by aberrant cryptic splicing due to TDP-43 nuclear loss can act as neo-antigens. We detect the HDGFL2 cryptic peptide and multiple other TDP-43 cryptic exons in IBM skeletal muscle, where their presence correlates with enrichment of T cells and class I antigen presentation pathways. Furthermore, we identify epitopes deriving from HDGFL2 and IGLON5 cryptic peptides which are recognized by clonally expanded and functionally differentiated populations of CD8 + T cells in ALS and IBM Patients. Finally, we demonstrate that T cells engineered to express the identified TCRs can bind and activate in response to the cryptic peptide derived epitopes (cryptic epitopes) and are able to kill TDP-43 deficient astrocytes. This work identifies for the first time specific T cell antigens in ALS and IBM, directly linking adaptive immune response to TDP-43 pathology.

Introduction Patients (pts) with multiple myeloma (MM) eventually become refractory to current treatments. There is an unmet need for new therapies with durable efficacy, favorable safety, and simplified dosing schedules. ABBV-383 is a unique BCMA x CD3 bispecific antibody T-cell engager comprising bivalent high-avidity BCMA binding domains, a low-affinity CD3-binding domain designed to minimize cytokine release and the risk for cytokine release syndrome (CRS), and a silenced Fc tail for extended half-life enabling dosing convenience. In an ongoing first-in-human (FIH) study (NCT03933735) in heavily pretreated pts with relapsed or refractory (RR) MM, ABBV-383 monotherapy (60 mg Q4W) resulted in deep and durable responses. Introduction of a modified dexamethasone (Dex) premedication (premed) schedule in cycle (C) 1 lowered incidence and severity of CRS (43% overall; 5% grade ≥2) vs pts treated with low Dex (71% overall; 20% grade ≥2) (JCO 2024;42[suppl 16]:7531). We now report results of Arm A of the open-label, phase 1b study (NCT05650632) evaluating 1 step-up dose (SUD) of ABBV-383 as a strategy to mitigate the risk of severe CRS and further assess ABBV-383 activity in pts with RRMM. Methods This study is enrolling pts with RRMM, ECOG performance status ≤2, and documented evidence of progression during or after the last treatment. Pts must have received ≥3 prior lines of therapy, including a proteasome inhibitor, an immunomodulatory drug, and an anti-CD38 monoclonal antibody. Arm A consists of a dose-optimization (DO) and a dose-expansion (DE) part. During DO, pts received a single SUD of intravenous ABBV-383 (2/4 mg) on C1 day (D) 1, followed by the target dose of 60 mg on D4. In the DE, based on DO findings, pts received the recommended ABBV-383 SUD on C1D1, followed by 60 mg on D4. The modified premed schedule that lowered incidence and severity of CRS in the FIH study was also introduced prior to target dose on C1D4. After C1, pts received 60 mg ABBV-383 Q4W on D1 of each cycle for DO and DE until disease progression, withdrawal, or discontinuation. Primary endpoint was grade ≥2 CRS events during C1. Other endpoints included any grade CRS events, treatment-emergent adverse events (TEAEs), immune pharmacodynamics, and clinical activity. Results As of April 2024, 70 pts were enrolled in Arm A: 47 in DO (2 mg SUD, n=26; 4 mg SUD, n=21) and 23 in DE. Median age was 69 years (range: 40-84), 59% were male, and 76% were triple-class refractory. Median prior lines of therapy was 4 (range: 3-10). Median follow-up time (months [mo]) was 6.5, 7.5, and 2.8 for 2 mg DO, 4 mg DO, and DE, respectively. Any grade CRS was reported in 10 (39%) and 11 (52%) pts during DO at 2 and 4 mg SUD; CRS grade ≥2 occurred in 3 (12%) pts in the 2 mg SUD and in 6 (29%) pts in the 4 mg SUD, leading to selection of 2 mg SUD for DE. In the DE, 7 (30%) pts experienced CRS; 1 (4%) pt experienced grade 2 CRS, there were no grade ≥3 events, and only 2 (9%) pts received tocilizumab to treat CRS. Overall, median time to CRS onset and resolution was 14.7 and 9 hours, respectively; no pts had CRS after C1. Immune effector cell-associated neurotoxicity syndrome occurred in 7 (10%) pts (grade 1: 4%; grade 2: 6%; grade 3: 4%). Most common TEAEs were CRS (40%), neutropenia (37%), diarrhea (29%), anemia (24%), and fatigue (20%); most common grade 3/4 TEAEs were neutropenia (31%), anemia (19%), thrombocytopenia (13%), and leukopenia (11%). Grade 5 TEAEs occurred in 5 pts; 1 event was deemed possibly related to ABBV-383 (COVID-19 pneumonia). In the efficacy-evaluable population (n=68), the objective response rate was 62% (95% CI: 49.2, 73.3); 35 (52%) pts had a VGPR or better. Median time to response was 1.1 mo (range: 1-4) and median duration of follow-up was 5.8 mo (range: 1-12). Response rates are expected to improve with longer follow-up. Maximum reduction in peak levels of CRS-related cytokines, including IL-6, were observed in DE compared with DO and 60 mg Q4W without SUD (FIH study). Peak activation and proliferation of CD8 T cells was comparable with ABBV-383 in DO, DE, and 60 mg Q4W-treated pts. Conclusions Introduction of 1 SUD of 2 mg on D1 followed by full dose of 60 mg on D4 in C1 combined with modification to the Dex premed schedule reduced the incidence and severity of CRS in pts with RRMM, further optimizing the safety profile of ABBV-383. Preliminary efficacy data show early and high response rates, consistent with outcomes from other ABBV-383 studies.

Abstract Mass spectrometry (MS) is a technique widely employed for the identification and characterization of proteins, with personalized medicine, systems biology, and biomedical applications. The application of MS-based proteomics advances our understanding of protein function, cellular signaling, and complex biological systems. MS data analysis is a critical process that includes identifying and quantifying proteins and peptides and then exploring their biological functions in downstream analyses. To address the complexities associated with MS data analysis, we developed ProtPipe to streamline and automate the processing and analysis of high-throughput proteomics and peptidomics datasets with DIA-NN preinstalled. The pipeline facilitates data quality control, sample filtering, and normalization, ensuring robust and reliable downstream analyses. ProtPipe provides downstream analyses, including protein and peptide differential abundance identification, pathway enrichment analysis, protein–protein interaction analysis, and major histocompatibility complex (MHC)–peptide binding affinity analysis. ProtPipe generates annotated tables and visualizations by performing statistical post-processing and calculating fold changes between predefined pairwise conditions in an experimental design. It is an open-source, well-documented tool available at https://github.com/NIH-CARD/ProtPipe, with a user-friendly web interface.