Chiraag Gohel https://www.my-great-forest.net/index/ index /index/ Chiraag Gohel Person false https://chiraaggohel.com Ph.D. Student I am a Ph.D. candidate in Biostatistics and Bioinformatics at The George Washington University, advised by Dr. Ali Rahnavard. My research focuses on statistical and machine learning methods for high-dimensional multi-omics data, with applications in metabolomics.You can email me at chiraaggohel AT gwu DOT com. CodeElliot PickensChiraag GohelSidharth Satyahttps://www.my-great-forest.net/pfn-npe/pfn-npe/pfn-npe/PFN-NPERepositoryhttps://github.com/epickens/pfn-npePythonResearch code for PFN-NPE, using frozen TabPFN representations as summary statistics for simulation-based inference.Repository: GitHub.Related publication: Pre-Trained Tabular Foundation Models as Versatile Summary Networks for Neural Posterior Estimation. References Elliot Pickens Chiraag Gohel Sidharth Satya 2026 5 8 https://www.my-great-forest.net/pub-pickens-pretrained-tabular-foundation-2026/ pub-pickens-pretrained-tabular-foundation-2026 /pub-pickens-pretrained-tabular-foundation-2026/ Pre-Trained Tabular Foundation Models as Versatile Summary Networks for Neural Posterior Estimation Reference pickensPretrainedTabularFoundation2026 bib.bib pub-pickens-pretrained-tabular-foundation-2026 10.48550/arXiv.2605.07765 http://arxiv.org/abs/2605.07765 Abstract. In this work, we study TabPFN as a training-free, modular summary network for simulation-based Bayesian inference (SBI). Tabular foundation models such as TabPFN are pretrained on broad families of synthetic tabular data-generating processes and adapt at test time through in-context learning, making them natural candidates for SBI, where posterior estimation often depends on learning informative summaries of simulated observations. We propose PFN-NPE: a general recipe that uses a pretrained TabPFN encoder as a fixed summary network for simulator outputs, then pairs the resulting summaries with a downstream inference head chosen for the problem. With normalizing flows as the default inference head, PFN-NPE matches established posterior approximation methods and sometimes outperforms them. More importantly, diagnostic probes show that the TabPFN-derived summaries often preserve useful posterior location and marginal information. These analyses also reveal a limitation in that TabPFN-derived summaries may struggle to represent the joint posterior structure even when the marginals are well recovered. Still, our experiments show that TabPFN can serve as an effective summary network across a diverse set of SBI settings, with the inference network left modular and task-dependent. Context Backlinks Related Contributions Elliot Pickens Chiraag Gohel Sidharth Satya 2026 5 8 https://www.my-great-forest.net/pub-pickens-pretrained-tabular-foundation-2026/ pub-pickens-pretrained-tabular-foundation-2026 /pub-pickens-pretrained-tabular-foundation-2026/ Pre-Trained Tabular Foundation Models as Versatile Summary Networks for Neural Posterior Estimation Reference pickensPretrainedTabularFoundation2026 bib.bib pub-pickens-pretrained-tabular-foundation-2026 10.48550/arXiv.2605.07765 http://arxiv.org/abs/2605.07765 Abstract. In this work, we study TabPFN as a training-free, modular summary network for simulation-based Bayesian inference (SBI). Tabular foundation models such as TabPFN are pretrained on broad families of synthetic tabular data-generating processes and adapt at test time through in-context learning, making them natural candidates for SBI, where posterior estimation often depends on learning informative summaries of simulated observations. We propose PFN-NPE: a general recipe that uses a pretrained TabPFN encoder as a fixed summary network for simulator outputs, then pairs the resulting summaries with a downstream inference head chosen for the problem. With normalizing flows as the default inference head, PFN-NPE matches established posterior approximation methods and sometimes outperforms them. More importantly, diagnostic probes show that the TabPFN-derived summaries often preserve useful posterior location and marginal information. These analyses also reveal a limitation in that TabPFN-derived summaries may struggle to represent the joint posterior structure even when the marginals are well recovered. Still, our experiments show that TabPFN can serve as an effective summary network across a diverse set of SBI settings, with the inference network left modular and task-dependent. Jennie Caldwell Krunal Parekh Brandon Crowther Chiraag Gohel Roberta Pileggi A. Isabel Garcia Mina Ghorbanifarajzadeh Teresa A. Dolan Anita Gohel 2026 1 29 https://www.my-great-forest.net/pub-caldwell-performance-evaluation-ai-based-2026/ pub-caldwell-performance-evaluation-ai-based-2026 /pub-caldwell-performance-evaluation-ai-based-2026/ Performance Evaluation of AI-based Caries Detection Technology and Its Educational Training Module: A Dual-Phase Investigation Reference caldwellPerformanceEvaluationAIbased2026 bib.bib pub-caldwell-performance-evaluation-ai-based-2026 Frontiers in Dental Medicine 10.3389/fdmed.2025.1741855 https://www.frontiersin.org/journals/dental-medicine/articles/10.3389/fdmed.2025.1741855/full Abstract. PurposeThis objective of the study was to assess the accuracy of an AI-based caries detection system, Overjet Caries Assist, or OCA, (Overjet Inc. Claymont, DE, USA) and to analyze the efficacy of the AI-based caries detection training module in teaching dental students radiographic diagnosis of dental caries.MethodsTwo independent calibrated observers evaluated 1604 proximal surfaces of teeth on intraoral bitewing radiographs and compared the findings to the OCA caries detection module. The sensitivity, specificity, positive and negative predictive values and diagnostic accuracy of the AI system were calculated. For the second part of the study, 82 first- and third-year dental students interpreted 10 intraoral bitewings for caries diagnosis before and after undergoing training with the OCA AI-caries detection training module. Non-parametric Wilcoxon signed-rank test was used to assess the difference between the students’ learning before and after the Overjet module training.ResultsThe average sensitivity for enamel lesions was found to be 0.69, while the average sensitivity for dentinal lesions was found to be 0.91. The average specificity for enamel lesions was found to be 0.99, while the average specificity for dentinal lesions was found to be 0.98. There was a 43 percent–46 percent increase in students’ ability to detect radiographic caries following the completion of the Overjet training module.ConclusionThe OCA module can be used as an effective tool for assisting diagnosis of caries among dental students. The Overjet training module is effective in training dental students’ radiographic diagnosis of caries. Ricky Amreek Suri Chiraag Gohel Wazeer Alghamdi Brandon Crowther A. Isabel Garcia Anita Gohel 2026 1 2 https://www.my-great-forest.net/pub-suri-perceptions-dental-students-2026/ pub-suri-perceptions-dental-students-2026 /pub-suri-perceptions-dental-students-2026/ Perceptions of Dental Students on the Integration of Artificial Intelligence in Radiology Clinical Education Reference suriPerceptionsDentalStudents2026 bib.bib pub-suri-perceptions-dental-students-2026 Frontiers in Dental Medicine 10.3389/fdmed.2025.1735299 https://www.frontiersin.org/journals/dental-medicine/articles/10.3389/fdmed.2025.1735299/full $} 0.80), however, significant negative correlations emerged between student outcomes and perceptions of faculty preparedness.ConclusionsStudents value AI as a diagnostic learning aid but identify gaps in curricular structure and faculty calibration. A structured, faculty-led AI curriculum introduced early in training and paired with patient communication strategies may optimize preparedness while safeguarding critical thinking.}, langid = {english}, keywords = {artificial intelligence,automation bias,curriculum development,dental education,radiology} }]]> Abstract. ObjectiveTo assess dental students' perceptions of artificial intelligence (AI) in radiology education, focusing on diagnostic value, curriculum preparedness, and faculty support.MethodsAn anonymous survey was administered to third-year dental students (n = 66, response rate 71.7 percent) at the University of Florida College of Dentistry after exposure to the Overjet Caries Assist (OCA) platform (Overjet Inc. Claymont, DE, USA). Likert-scale, multiple-choice, and open-ended items captured attitudes toward diagnostic accuracy, skill development, curriculum integration, and patient communication. Descriptive statistics, polychoric correlations with bootstrap resampling, and thematic analysis of qualitative responses were conducted.ResultsMost students reported that AI improved their ability to detect caries (89.4 percent) and enhanced radiographic interpretation (92.4 percent). However, only 16.7 percent agreed the curriculum adequately prepared them to use AI clinically, and just 45.5 percent felt confident about integrating AI into future practice. Open-ended feedback highlighted three themes: 1. need for structured faculty training, 2. earlier and more frequent AI exposure, and 3. emphasis on mitigating automation bias, or the over reliance on technology and automated systems in clinical judgement. Correlation analysis revealed strong associations between improved interpretation, skill development, and patient communication (r > 0.80), however, significant negative correlations emerged between student outcomes and perceptions of faculty preparedness.ConclusionsStudents value AI as a diagnostic learning aid but identify gaps in curricular structure and faculty calibration. A structured, faculty-led AI curriculum introduced early in training and paired with patient communication strategies may optimize preparedness while safeguarding critical thinking. Veronica M. Amuso MaryEllen R. Haas Paula O. Cooper Ranojoy Chatterjee Sana Hafiz Shatha Salameh Chiraag Gohel Miguel F. Mazumder Violet Josephson Sarah S. Kleb Khatereh Khorsandi Anelia Horvath Ali Rahnavard Brett A. Shook 2025 7 1 https://www.my-great-forest.net/pub-amuso-fibroblast-mediated-macrophage-recruitment-2025/ pub-amuso-fibroblast-mediated-macrophage-recruitment-2025 /pub-amuso-fibroblast-mediated-macrophage-recruitment-2025/ Fibroblast-Mediated Macrophage Recruitment Supports Acute Wound Healing Reference amusoFibroblastMediatedMacrophageRecruitment2025 bib.bib pub-amuso-fibroblast-mediated-macrophage-recruitment-2025 Journal of Investigative Dermatology 10.1016/j.jid.2024.10.609 https://www.sciencedirect.com/science/article/pii/S0022202X24029567 Abstract. Epithelial and immune cells have long been appreciated for their contribution to the early immune response after injury; however, much less is known about the role of mesenchymal cells. Using single-nuclei RNA sequencing, we defined changes in gene expression associated with inflammation 1 day after wounding in mouse skin. Compared with those in keratinocytes and myeloid cells, we detected enriched expression of proinflammatory genes in fibroblasts associated with deeper layers of the skin. In particular, SCA1+ fibroblasts were enriched for numerous chemokines, including CCL2, CCL7, and IL-33, compared with SCA1− fibroblasts. Genetic deletion of Ccl2 in fibroblasts resulted in fewer wound-bed macrophages and monocytes during injury-induced inflammation, with reduced revascularization and re-epithelialization during the proliferation phase of healing. These findings highlight the important contribution of fibroblast-derived factors to injury-induced inflammation and the impact of immune cell dysregulation on subsequent tissue repair. Andrew Patt Iris Pang Fred Lee Chiraag Gohel Eoin Fahy Vicki Stevens David Ruggieri Steven C. Moore Ewy A. Mathé 2025 5 2 https://www.my-great-forest.net/pub-patt-met-link-r-facilitating-metaanalysis-2025/ pub-patt-met-link-r-facilitating-metaanalysis-2025 /pub-patt-met-link-r-facilitating-metaanalysis-2025/ metLinkR: Facilitating Metaanalysis of Human Metabolomics Data through Automated Linking of Metabolite Identifiers Reference pattMetLinkRFacilitatingMetaanalysis2025 bib.bib pub-patt-met-link-r-facilitating-metaanalysis-2025 Journal of Proteome Research 10.1021/acs.jproteome.4c01051 https://doi.org/10.1021/acs.jproteome.4c01051 Abstract. Metabolites are referenced in spectral, structural and pathway databases with a diverse array of schemas, including various internal database identifiers and large tables of common name synonyms. Cross-linking metabolite identifiers is a required step for meta-analysis of metabolomic results across studies but made difficult due to the lack of a consensus identifier system. We have implemented metLinkR, an R package that leverages RefMet and RaMP-DB to automate and simplify cross-linking metabolite identifiers across studies and generating common names. MetLinkR accepts as input metabolite common names and identifiers from five different databases (HMDB, KEGG, ChEBI, LIPIDMAPS and PubChem) to exhaustively search for possible overlap in supplied metabolites from input data sets. In an example of 13 metabolomic data sets totaling 10,400 metabolites, metLinkR identified and provided common names for 1377 metabolites in common between at least 2 data sets in less than 18 min and produced standardized names for 74.4 percent of the input metabolites. In another example comprising five data sets with 3512 metabolites, metLinkR identified 715 metabolites in common between at least two data sets in under 12 min and produced standardized names for 82.3 percent of the input metabolites. Outputs of MetLInR include output tables and metrics allowing users to readily double check the mappings and to get an overview of chemical classes represented. Overall, MetLinkR provides a streamlined solution for a common task in metabolomic epidemiology and other fields that meta-analyze metabolomic data. The R package, vignette and source code are freely downloadable at https://github.com/ncats/metLinkR. Mackenzie E. Smith Chuck T. Chen Chiraag Gohel Giulia Cisbani Daniel K. Chen Kimia Rezaei Andrew McCutcheon Richard P. Bazinet 2024 1 17 https://www.my-great-forest.net/pub-smith-upregulated-hepatic-lipogenesis-2024/ pub-smith-upregulated-hepatic-lipogenesis-2024 /pub-smith-upregulated-hepatic-lipogenesis-2024/ Upregulated Hepatic Lipogenesis from Dietary Sugars in Response to Low Palmitate Feeding Supplies Brain Palmitate Reference smithUpregulatedHepaticLipogenesis2024 bib.bib pub-smith-upregulated-hepatic-lipogenesis-2024 Nature Communications 10.1038/s41467-023-44388-4 https://www.nature.com/articles/s41467-023-44388-4 Abstract. Palmitic acid (PAM) can be provided in the diet or synthesized via de novo lipogenesis (DNL), primarily, from glucose. Preclinical work on the origin of brain PAM during development is scarce and contrasts results in adults. In this work, we use naturally occurring carbon isotope ratios (13C/12C; δ13C) to uncover the origin of brain PAM at postnatal days 0, 10, 21 and 35, and RNA sequencing to identify the pathways involved in maintaining brain PAM, at day 35, in mice fed diets with low, medium, and high PAM from birth. Here we show that DNL from dietary sugars maintains the majority of brain PAM during development and is augmented in mice fed low PAM. Importantly, the upregulation of hepatic DNL genes, in response to low PAM at day 35, demonstrates the presence of a compensatory mechanism to maintain total brain PAM pools compared to the liver; suggesting the importance of brain PAM regulation. Claire Gao Chiraag Gohel Yan Leng Jun Ma David Goldman Ariel J Levine Mario A Penzo 2023 3 3 https://www.my-great-forest.net/pub-gao-molecular-spatial-profiling-2023/ pub-gao-molecular-spatial-profiling-2023 /pub-gao-molecular-spatial-profiling-2023/ Molecular and Spatial Profiling of the Paraventricular Nucleus of the Thalamus Reference gaoMolecularSpatialProfiling2023 bib.bib pub-gao-molecular-spatial-profiling-2023 eLife 10.7554/eLife.81818 https://doi.org/10.7554/eLife.81818 Abstract. The paraventricular nucleus of the thalamus (PVT) is known to regulate various cognitive and behavioral processes. However, while functional diversity among PVT circuits has often been linked to cellular differences, the molecular identity and spatial distribution of PVT cell types remain unclear. To address this gap, here we used single nucleus RNA sequencing (snRNA-seq) and identified five molecularly distinct PVT neuronal subtypes in the mouse brain. Additionally, multiplex fluorescent in situ hybridization of top marker genes revealed that PVT subtypes are organized by a combination of previously unidentified molecular gradients. Lastly, comparing our dataset with a recently published single-cell sequencing atlas of the thalamus yielded novel insight into the PVT’s connectivity with the cortex, including unexpected innervation of auditory and visual areas. This comparison also revealed that our data contains a largely non-overlapping transcriptomic map of multiple midline thalamic nuclei. Collectively, our findings uncover previously unknown features of the molecular diversity and anatomical organization of the PVT and provide a valuable resource for future investigations. Danielle Sambo Chiraag Gohel Qiaoping Yuan Gauthaman Sukumar Camille Alba Clifton L. Dalgard David Goldman 2022 https://www.my-great-forest.net/pub-sambo-cell-typespecific-changes-2022/ pub-sambo-cell-typespecific-changes-2022 /pub-sambo-cell-typespecific-changes-2022/ Cell Type-Specific Changes in Wnt Signaling and Neuronal Differentiation in the Developing Mouse Cortex after Prenatal Alcohol Exposure during Neurogenesis Reference samboCellTypespecificChanges2022 bib.bib pub-sambo-cell-typespecific-changes-2022 Frontiers in Cell and Developmental Biology 10.3389/fcell.2022.1011974 https://doi.org/10.3389/fcell.2022.1011974 Abstract. Fetal Alcohol Spectrum Disorder (FASD) encompasses an array of effects of prenatal alcohol exposure (PAE), including physical abnormalities and cognitive and behavioral deficits. Disruptions of cortical development have been implicated in multiple PAE studies, with deficits including decreased progenitor proliferation, disrupted neuronal differentiation, aberrant radial migration of pyramidal neurons, and decreased cortical thickness. While several mechanisms of alcohol teratogenicity have been explored, how specific cell types in the brain at different developmental time points may be differentially affected by PAE is still poorly understood. In this study, we used single nucleus RNA sequencing (snRNAseq) to investigate whether moderate PAE from neurulation through peak cortical neurogenesis induces cell type-specific transcriptomic changes in the developing murine brain. Cluster analysis identified 25 neuronal cell types, including subtypes of radial glial cells (RGCs), intermediate progenitor cells (IPCs), projection neurons, and interneurons. Only Wnt-expressing cortical hem RGCs showed a significant decrease in the percentage of cells after PAE, with no cell types showing PAE-induced apoptosis as measured by caspase expression. Cell cycle analysis revealed only a subtype of RGCs expressing the downstream Wnt signaling transcription factor Tcf7l2 had a decreased percentage of cells in the G2/M phase of the cell cycle, suggesting decreased proliferation in this RGC subtype and further implicating disrupted Wnt signaling after PAE at this early developmental timepoint. An increased pseudotime score in IPC and projection neuron cell types indicated that PAE led to increased or premature differentiation of these cells. Biological processes affected by PAE included the upregulation of pathways related to synaptic activity and neuronal differentiation and downregulation of pathways related to chromosome structure and the cell cycle. Several cell types showed a decrease in Wnt signaling pathways, with several genes related to Wnt signaling altered by PAE in multiple cell types. As Wnt has been shown to promote proliferation and inhibit differentiation at earlier stages in development, the downregulation of Wnt signaling may have resulted in premature neuronal maturation of projection neurons and their intermediate progenitors. Overall, these findings provide further insight into the cell type-specific effects of PAE during early corticogenesis.