Nelson B. Freimer
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Nelson B. Freimer has a total of 1000 co-authors. Here are the top co-authors by number of shared publications:
Chiara Sabatti
Department of Biomedical Data Sciences, Stanford University, Stanford, CA, USA
21 shared publicationsSusan K. Service
Center for Neurobehavioral Genetics, Jane and Terry Semel Institute for Neuroscience and Human Behavior, University of California Los Angeles, Los Angeles, USA
21 shared publications- Victor I. Reus
UCSF
14 shared publications Aarno Palotie
Institute for Molecular Medicine Finland (FIMM), University of Helsinki, Helsinki, Finland
13 shared publicationsJulio Molina
Department of Psychiatry and Biobehavioral Science, University of California, Los Angeles, CA, USA
11 shared publications- Matthew Jorgensen
Wake Forest University School of Medicine, Winston‐Salem, NC USA
11 shared publications
Latest Publications
- Open Access
- 26 October 2026
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 28, e93328
Generative Large Language Models for Psychiatric Concept Detection in Spanish-Language Electronic Health Records: Development and Evaluation Study
Background: The accurate detection of clinical symptoms in electronic health records (EHRs) is pivotal for advancing large-scale genetic and longitudinal studies in psychiatry. Free-text clinical notes are an essential source of symptom-level information, particularly in psychiatry; however, the automated extraction of symptoms from clinical text remains challenging. Objective: We aimed to evaluate the ability of open-source generative large language models (LLMs) to detect psychiatric symptom-level concepts from clinical text in Spanish and to develop a new fine-tuned, open-source LLM that can be shared without risk of exposing patient information. Methods: We tested 11 open-source generative LLMs for their ability to detect 109 psychiatric concepts from clinical text, using annotated EHR notes from a psychiatric hospital in Colombia. The LLMs were evaluated both “out-of-the-box” and after fine-tuning and compared against a traditional natural language processing (tNLP) method developed from the same data. To develop a fine-tuned LLM that can be shared without risk of exposing patient information, we created a novel synthetic dataset, leveraging real concept annotations while generating fully synthetic clinical text, and fine-tuned a top-performing LLM on these data. Fine-tuned model generalizability was assessed through cross-site evaluation using data from a second independent hospital in Colombia. Results: We show that while base LLM performance was poor to moderate (zero-shot: 0.2-0.6 macro-F1; few-shot: 0.2-0.74 macro-F1), it improved significantly after fine-tuning (0.75-0.86 macro-F1), with several fine-tuned LLMs outperforming the tNLP method. In total, 100 psychiatric concepts could be reliably detected (F1>0.8) using either a fine-tuned LLM or tNLP. Using the synthetic dataset, we developed Mistral-small-psych, a new LLM capable of detecting psychiatric concepts from Spanish clinical text with performance comparable to models trained on real EHR data (macro-F1=0.79) while remaining fully distributable. In cross-site evaluations, most LLMs generalized well, with small out-of-domain performance losses (macro-F1 decrease of 0.02-0.16). Conclusions: This work provides the first large-scale, cross-institutional evaluation of state-of-the-art LLMs for psychiatric concept detection in Spanish EHRs, introduces a privacy-preserving pathway for building shareable clinical natural language processing models, and presents a new open model for extracting symptom-level psychiatric information from Spanish clinical text.
- 26 October 2026
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 28, e93328
Open AccessGenerative Large Language Models for Psychiatric Concept Detection in Spanish-Language Electronic Health Records: Development and Evaluation Study
Background: The accurate detection of clinical symptoms in electronic health records (EHRs) is pivotal for advancing large-scale genetic and longitudinal studies in psychiatry. Free-text clinical notes are an essential source of symptom-level information, particularly in psychiatry; however, the automated extraction of symptoms from clinical text remains challenging. Objective: We aimed to evaluate the ability of open-source generative large language models (LLMs) to detect psychiatric symptom-level concepts from clinical text in Spanish and to develop a new fine-tuned, open-source LLM that can be shared without risk of exposing patient information. Methods: We tested 11 open-source generative LLMs for their ability to detect 109 psychiatric concepts from clinical text, using annotated EHR notes from a psychiatric hospital in Colombia. The LLMs were evaluated both “out-of-the-box” and after fine-tuning and compared against a traditional natural language processing (tNLP) method developed from the same data. To develop a fine-tuned LLM that can be shared without risk of exposing patient information, we created a novel synthetic dataset, leveraging real concept annotations while generating fully synthetic clinical text, and fine-tuned a top-performing LLM on these data. Fine-tuned model generalizability was assessed through cross-site evaluation using data from a second independent hospital in Colombia. Results: We show that while base LLM performance was poor to moderate (zero-shot: 0.2-0.6 macro-F1; few-shot: 0.2-0.74 macro-F1), it improved significantly after fine-tuning (0.75-0.86 macro-F1), with several fine-tuned LLMs outperforming the tNLP method. In total, 100 psychiatric concepts could be reliably detected (F1>0.8) using either a fine-tuned LLM or tNLP. Using the synthetic dataset, we developed Mistral-small-psych, a new LLM capable of detecting psychiatric concepts from Spanish clinical text with performance comparable to models trained on real EHR data (macro-F1=0.79) while remaining fully distributable. In cross-site evaluations, most LLMs generalized well, with small out-of-domain performance losses (macro-F1 decrease of 0.02-0.16). Conclusions: This work provides the first large-scale, cross-institutional evaluation of state-of-the-art LLMs for psychiatric concept detection in Spanish EHRs, introduces a privacy-preserving pathway for building shareable clinical natural language processing models, and presents a new open model for extracting symptom-level psychiatric information from Spanish clinical text.
- Open Access
A founder population reveals genetic variation with major effect on serious mental illness
Genetic studies of serious mental illness (SMI, including schizophrenia and bipolar disorder) have implicated two classes of risk variant: common variants 1–3 , each of small effect but collectively accounting for most population-level risk, and ultra-rare coding variants 4,5 which confer large risk in a small number of individuals. Most such studies cannot directly assess the impact of specific variants on SMI, transdiagnostically, as the phenotypic information that they collect is focused on a single diagnosis. Here we report results of association analysis of predicted deleterious variants from exome sequencing of 21,958 cases (SMI and severe major depressive disorder) and 19,826 controls from the Mision Origen biobank in the Paisa genetic isolate of Colombia. We identified transdiagnostic associations to seven genes, three at an exome-wide and four at a false discovery rate significance threshold. Five of these genes have not previously been implicated in SMI ( ADAP1 , WIPI1 , NFE2L1 , FBXO10 , and HLA-DPB1 ); for the first four of these the association is driven by variants enriched in the Paisa due to a founder effect, filling the gap between ultra-rare and common variants. The strongest association that we observe is to a frameshift variant (7:904169:T:TC) in ADAP1 that is genome-wide significant for SMI, schizophrenia, and bipolar disorder and increases risk for both diagnoses by more than threefold; it truncates ADAP1, a brain-enriched Arf6 GTPase-activating/scaffolding protein that has been mechanistically linked to post-mitotic neuronal morphogenesis and regulation of dendritic differentiation 6 . Variant carriers differ significantly from other SMI cases in Mision Origen for psychotic, manic, and cognitive item-level phenotypes assessed through analyses of electronic health records 7–9 and displayed impaired executive function on neurocognitive testing. Variant carriers differ significantly from other SMI cases in Mision Origen for multiple item-level phenotypes (psychotic, manic, and cognitive) assigned through natural language processing of longitudinal electronic health records. Although the 7:904169:T:TC variant is not significantly associated with major depressive disorder, carriers compared to others with this diagnosis display a significant increase in symptoms typical of bipolar disorder. The variant’s high allele frequency in Mision Origen (∼0.4%) will enable population-level studies to stratify SMI based on causation and identify additional factors influencing risk and resilience.
Open AccessA founder population reveals genetic variation with major effect on serious mental illness
Genetic studies of serious mental illness (SMI, including schizophrenia and bipolar disorder) have implicated two classes of risk variant: common variants 1–3 , each of small effect but collectively accounting for most population-level risk, and ultra-rare coding variants 4,5 which confer large risk in a small number of individuals. Most such studies cannot directly assess the impact of specific variants on SMI, transdiagnostically, as the phenotypic information that they collect is focused on a single diagnosis. Here we report results of association analysis of predicted deleterious variants from exome sequencing of 21,958 cases (SMI and severe major depressive disorder) and 19,826 controls from the Mision Origen biobank in the Paisa genetic isolate of Colombia. We identified transdiagnostic associations to seven genes, three at an exome-wide and four at a false discovery rate significance threshold. Five of these genes have not previously been implicated in SMI ( ADAP1 , WIPI1 , NFE2L1 , FBXO10 , and HLA-DPB1 ); for the first four of these the association is driven by variants enriched in the Paisa due to a founder effect, filling the gap between ultra-rare and common variants. The strongest association that we observe is to a frameshift variant (7:904169:T:TC) in ADAP1 that is genome-wide significant for SMI, schizophrenia, and bipolar disorder and increases risk for both diagnoses by more than threefold; it truncates ADAP1, a brain-enriched Arf6 GTPase-activating/scaffolding protein that has been mechanistically linked to post-mitotic neuronal morphogenesis and regulation of dendritic differentiation 6 . Variant carriers differ significantly from other SMI cases in Mision Origen for psychotic, manic, and cognitive item-level phenotypes assessed through analyses of electronic health records 7–9 and displayed impaired executive function on neurocognitive testing. Variant carriers differ significantly from other SMI cases in Mision Origen for multiple item-level phenotypes (psychotic, manic, and cognitive) assigned through natural language processing of longitudinal electronic health records. Although the 7:904169:T:TC variant is not significantly associated with major depressive disorder, carriers compared to others with this diagnosis display a significant increase in symptoms typical of bipolar disorder. The variant’s high allele frequency in Mision Origen (∼0.4%) will enable population-level studies to stratify SMI based on causation and identify additional factors influencing risk and resilience.
- Open Access
Large-Scale Psychiatric Concept Extraction from Electronic Health Records: A Comparative Study of Encoder-Based Language Models
Background: Free-text notes in electronic health records (EHRs) contain fine-grained psychiatric information that is essential for psychiatric research and clinical care, and often absent or under-recorded in structured codes alone. Clinical natural language processing (cNLP) can support extraction of this information from EHR notes, yet Spanish-language cNLP remains under-developed. Moreover, broad evaluations comparing multiple encoder-based language models across extensive, fine-grained psychiatric concept sets remain scarce, and it remains unclear how these models compare with traditional NLP (tNLP) systems and much larger generative large language models (LLMs). In addition, cross-site performance of fine-tuned models is rarely tested, and limited annotated training data remains a major challenge, especially for rare symptoms. Objectives: We aimed to advance scalable, global psychiatric cNLP by fine-tuning multiple encoder-based models with differing architectures and pre-training strategies for detecting fine-grained psychiatric concepts in Spanish EHRs. We further evaluated the impact of augmenting the fine-tuning data with precision-weighted weak labels for less-frequent concepts, and compared the performance of the encoder-based models to that of tNLP and a fine-tuned generative LLM trained on the same data. Finally, we evaluated model cross-site generalizability on an external EHR dataset. Methods: Three encoder-based models (BETO, XLM-RoBERTa-large, and bsc-bio-ehr-es) were fine-tuned on 1,642 clinician-annotated EHR documents from Colombia to detect 110 psychiatric concepts in Spanish text. To address the limited annotated examples available for less-frequent concepts, 12,000 additional documents were weakly-labeled for less-frequent concepts using tNLP, and incorporated into the fine-tuning data with labels weighted by pattern precision. Models were compared with tNLP and a generative LLM, and evaluated on an external EHR dataset from another psychiatric hospital in Colombia. Results: Encoder model performance varied substantially, with macro-F1 ranging from 0.64 to 0.81. BETO achieved the highest macro-F1 (0.81; median F1=0.88 [IQR=0.77-0.96]). Adding precision-weighted weak labels for less-frequent concepts improved BETO’s overall macro-F1 to 0.83 and increased mean F1 for the 55 augmented concepts from 0.82 to 0.86. Under matched fine-tuning conditions, fine-tuned BETO and the tNLP method were equivalent in F1, whereas the LLM significantly outperformed BETO in F1. After weak-label augmentation, BETO significantly outperformed tNLP in F1 ( P FDR <.001) and narrowed the performance gap with the LLM, although equivalence was not established. Lastly, fine-tuned BETO maintained reasonably strong performance on data from an external hospital not used for model fine-tuning (out-of-domain macro-F1=0.78). Conclusions: General-purpose pre-trained encoders had strong performance for psychiatric concept extraction from Spanish EHRs. Weak-label augmentation improved BETO’s performance and strengthened results relative to a tNLP baseline, while reducing, but not eliminating, the performance gap with a much larger fine-tuned generative LLM. These findings highlight the utility of these relatively lightweight models for scalable, accurate and reproducible detection of psychiatric concepts in Spanish-language EHRs.
Open AccessLarge-Scale Psychiatric Concept Extraction from Electronic Health Records: A Comparative Study of Encoder-Based Language Models
Background: Free-text notes in electronic health records (EHRs) contain fine-grained psychiatric information that is essential for psychiatric research and clinical care, and often absent or under-recorded in structured codes alone. Clinical natural language processing (cNLP) can support extraction of this information from EHR notes, yet Spanish-language cNLP remains under-developed. Moreover, broad evaluations comparing multiple encoder-based language models across extensive, fine-grained psychiatric concept sets remain scarce, and it remains unclear how these models compare with traditional NLP (tNLP) systems and much larger generative large language models (LLMs). In addition, cross-site performance of fine-tuned models is rarely tested, and limited annotated training data remains a major challenge, especially for rare symptoms. Objectives: We aimed to advance scalable, global psychiatric cNLP by fine-tuning multiple encoder-based models with differing architectures and pre-training strategies for detecting fine-grained psychiatric concepts in Spanish EHRs. We further evaluated the impact of augmenting the fine-tuning data with precision-weighted weak labels for less-frequent concepts, and compared the performance of the encoder-based models to that of tNLP and a fine-tuned generative LLM trained on the same data. Finally, we evaluated model cross-site generalizability on an external EHR dataset. Methods: Three encoder-based models (BETO, XLM-RoBERTa-large, and bsc-bio-ehr-es) were fine-tuned on 1,642 clinician-annotated EHR documents from Colombia to detect 110 psychiatric concepts in Spanish text. To address the limited annotated examples available for less-frequent concepts, 12,000 additional documents were weakly-labeled for less-frequent concepts using tNLP, and incorporated into the fine-tuning data with labels weighted by pattern precision. Models were compared with tNLP and a generative LLM, and evaluated on an external EHR dataset from another psychiatric hospital in Colombia. Results: Encoder model performance varied substantially, with macro-F1 ranging from 0.64 to 0.81. BETO achieved the highest macro-F1 (0.81; median F1=0.88 [IQR=0.77-0.96]). Adding precision-weighted weak labels for less-frequent concepts improved BETO’s overall macro-F1 to 0.83 and increased mean F1 for the 55 augmented concepts from 0.82 to 0.86. Under matched fine-tuning conditions, fine-tuned BETO and the tNLP method were equivalent in F1, whereas the LLM significantly outperformed BETO in F1. After weak-label augmentation, BETO significantly outperformed tNLP in F1 ( P FDR <.001) and narrowed the performance gap with the LLM, although equivalence was not established. Lastly, fine-tuned BETO maintained reasonably strong performance on data from an external hospital not used for model fine-tuning (out-of-domain macro-F1=0.78). Conclusions: General-purpose pre-trained encoders had strong performance for psychiatric concept extraction from Spanish EHRs. Weak-label augmentation improved BETO’s performance and strengthened results relative to a tNLP baseline, while reducing, but not eliminating, the performance gap with a much larger fine-tuned generative LLM. These findings highlight the utility of these relatively lightweight models for scalable, accurate and reproducible detection of psychiatric concepts in Spanish-language EHRs.
- Open Access
- 4 August 2026
- Published by Springer Nature in npj Digital Public Health
- Vol. 1 (1) , 27
The Digital Mental Health Study assessed feasibility of large-scale digital sensing for depression and anxiety
Data passively obtained from smartphones and wearables can provide nearly continuous objective information that enables quantification of states and traits across broad physiological, behavioral, and emotional domains impacted in mental health conditions, including depression and anxiety. Widespread application of such digital phenotyping could transform the assessment of depression and anxiety in research and clinical care, but the field has lacked well-powered longitudinal studies demonstrating the utility of this approach. This paper describes the design and implementation of the Digital Mental Health Study (DMHS), which collected up to 12 months of sensor data from iPhone and Apple Watch in over 4000 consenting participants, a sample diverse by age, sex at birth, ethnicity, and depression symptom severity. To enable the use of these digital phenotypes to assay the complexity and heterogeneity of depression and anxiety, we designed a protocol of periodic self-report and interview-based scales optimized to assess elements of depression, anxiety, and perceived stress as broadly as possible while minimizing participant measurement burden. We report here the strategies used to recruit and enroll the DMHS sample, the process employed to develop study methods and protocols, and initial findings describing longitudinal symptom trajectories and demonstrating high participant engagement over 12 months.
- 4 August 2026
- Published by Springer Nature in npj Digital Public Health
- Vol. 1 (1) , 27
Open AccessThe Digital Mental Health Study assessed feasibility of large-scale digital sensing for depression and anxiety
Data passively obtained from smartphones and wearables can provide nearly continuous objective information that enables quantification of states and traits across broad physiological, behavioral, and emotional domains impacted in mental health conditions, including depression and anxiety. Widespread application of such digital phenotyping could transform the assessment of depression and anxiety in research and clinical care, but the field has lacked well-powered longitudinal studies demonstrating the utility of this approach. This paper describes the design and implementation of the Digital Mental Health Study (DMHS), which collected up to 12 months of sensor data from iPhone and Apple Watch in over 4000 consenting participants, a sample diverse by age, sex at birth, ethnicity, and depression symptom severity. To enable the use of these digital phenotypes to assay the complexity and heterogeneity of depression and anxiety, we designed a protocol of periodic self-report and interview-based scales optimized to assess elements of depression, anxiety, and perceived stress as broadly as possible while minimizing participant measurement burden. We report here the strategies used to recruit and enroll the DMHS sample, the process employed to develop study methods and protocols, and initial findings describing longitudinal symptom trajectories and demonstrating high participant engagement over 12 months.
- Open Access
- 8 July 2026
- Published by Springer Nature in Nature Genetics
- Vol. 58 (8) , 1813-1821
A blended genome and exome sequencing method captures genetic variation in an unbiased and cost-effective manner
Here we developed and deployed the blended genome exome (BGE) method, a DNA library approach that generates low-pass whole-genome (1–4× mean depth) and deep whole-exome (30–40× mean depth) data in a single sequencing run. BGE is cost-effective, empowers most genomic discoveries possible with deep whole-genome sequencing and captures global common single-nucleotide polymorphism diversity. We applied BGE to sequence >53,000 samples from the PUMAS Project (Populations Underrepresented in Mental Illness Associations Studies), including African, African American and Latin American populations. Imputed genotypes showed high concordance with Illumina Global Screening Array calls ( R 2 ≥ 95% for minor allele frequency ≥1%; ≥90% for minor allele frequency <1%), with consistent performance across local ancestries in admixed cohorts. For protein-coding copy number variants, deletions and duplications spanning at least three exons had a positive predicted value of ~90% relative to deep whole-genome data. At ~28% of the cost of deep whole-genome sequencing, BGE provides a scalable, reliable platform to expand genomic discovery and equitable access to sequencing in underrepresented populations.
- 8 July 2026
- Published by Springer Nature in Nature Genetics
- Vol. 58 (8) , 1813-1821
Open AccessA blended genome and exome sequencing method captures genetic variation in an unbiased and cost-effective manner
Here we developed and deployed the blended genome exome (BGE) method, a DNA library approach that generates low-pass whole-genome (1–4× mean depth) and deep whole-exome (30–40× mean depth) data in a single sequencing run. BGE is cost-effective, empowers most genomic discoveries possible with deep whole-genome sequencing and captures global common single-nucleotide polymorphism diversity. We applied BGE to sequence >53,000 samples from the PUMAS Project (Populations Underrepresented in Mental Illness Associations Studies), including African, African American and Latin American populations. Imputed genotypes showed high concordance with Illumina Global Screening Array calls ( R 2 ≥ 95% for minor allele frequency ≥1%; ≥90% for minor allele frequency <1%), with consistent performance across local ancestries in admixed cohorts. For protein-coding copy number variants, deletions and duplications spanning at least three exons had a positive predicted value of ~90% relative to deep whole-genome data. At ~28% of the cost of deep whole-genome sequencing, BGE provides a scalable, reliable platform to expand genomic discovery and equitable access to sequencing in underrepresented populations.
- Open Access
- 5 June 2026
- Published by Springer Nature in Research Square
Effect of ancestry and shared genetic architecture of serious mental illness on symptoms and cognition in an admixed Latin American population
Most genome-wide association studies (GWAS) of serious mental illness (SMI) have been conducted for categorical diagnoses in samples of primarily European ancestry. The portability of findings to non-Europeans, and to SMI-related symptoms/dimensional traits remains uncertain. In a sample of 8,666 SMI cases and controls from the Paisa region of Colombia we show that a primarily European schizophrenia GWAS polygenic risk score (PRS) predicted all SMI diagnoses in this sample, as well as symptoms (assessed in cases only) and traits assessed agnostic to SMI diagnosis: a one SD unit (SDU) increase in this PRS was associated to decreased risk in cases of suicidal thoughts (OR=0.89, 95% confidence interval 0.84-0.94), depressed mood (OR=0.90, 95% confidence interval 0.85-0.95), and increased risk of delusions (OR=1.12, 95% confidence interval 1.06-1.18) and to decreased cognition (in cases and controls) across five distinct domains (average decrease in cognition of 0.065 SDU, p<7e-05). We show that a published European GWAS of cognition predicted levels of executive functon(average decrease in cognition of 0.06 SDU per unit increase in PRS, p<2e-04), but not diagnosis or symptoms. Specific loci identified in the SMI GWAS also showed association to multiple diagnoses, symptoms, and cognitive traits in Paisa. The most noteworthy result was for a locus on chromosome 7p22.3, associated in multiple SMI GWAS, that showed association in Paisa to increased risk of bipolar disorder, and to reduced complex cognition and social cognition. Our findings demonstrate wide portability from European GWAS to an admixed American sample, with associations to multiple transdiagnostic phenotypes.
- 5 June 2026
- Published by Springer Nature in Research Square
Open AccessEffect of ancestry and shared genetic architecture of serious mental illness on symptoms and cognition in an admixed Latin American population
Most genome-wide association studies (GWAS) of serious mental illness (SMI) have been conducted for categorical diagnoses in samples of primarily European ancestry. The portability of findings to non-Europeans, and to SMI-related symptoms/dimensional traits remains uncertain. In a sample of 8,666 SMI cases and controls from the Paisa region of Colombia we show that a primarily European schizophrenia GWAS polygenic risk score (PRS) predicted all SMI diagnoses in this sample, as well as symptoms (assessed in cases only) and traits assessed agnostic to SMI diagnosis: a one SD unit (SDU) increase in this PRS was associated to decreased risk in cases of suicidal thoughts (OR=0.89, 95% confidence interval 0.84-0.94), depressed mood (OR=0.90, 95% confidence interval 0.85-0.95), and increased risk of delusions (OR=1.12, 95% confidence interval 1.06-1.18) and to decreased cognition (in cases and controls) across five distinct domains (average decrease in cognition of 0.065 SDU, p<7e-05). We show that a published European GWAS of cognition predicted levels of executive functon(average decrease in cognition of 0.06 SDU per unit increase in PRS, p<2e-04), but not diagnosis or symptoms. Specific loci identified in the SMI GWAS also showed association to multiple diagnoses, symptoms, and cognitive traits in Paisa. The most noteworthy result was for a locus on chromosome 7p22.3, associated in multiple SMI GWAS, that showed association in Paisa to increased risk of bipolar disorder, and to reduced complex cognition and social cognition. Our findings demonstrate wide portability from European GWAS to an admixed American sample, with associations to multiple transdiagnostic phenotypes.
- Open Access
Effect of ancestry and shared genetic architecture of serious mental illness on symptoms and cognition in an admixed Latin American population
Most genome-wide association studies (GWAS) of serious mental illness (SMI) have been conducted for categorical diagnoses in samples of primarily European ancestry. The portability of findings to non-Europeans, and to SMI-related symptoms/dimensional traits remains uncertain. In a sample of 8,666 SMI cases and controls from the Paisa region of Colombia we show that a primarily European schizophrenia GWAS polygenic risk score (PRS) predicted all SMI diagnoses in this sample, as well as symptoms (assessed in cases only) and traits assessed agnostic to SMI diagnosis: a one SD unit (SDU) increase in this PRS was associated to decreased risk in cases of suicidal thoughts (OR=0.89, 95% confidence interval 0.84-0.94), depressed mood (OR=0.90, 95% confidence interval 0.85-0.95), and increased risk of delusions (OR=1.12, 95% confidence interval 1.06-1.18) and to decreased cognition (in cases and controls) across five distinct domains (average decrease in cognition of 0.065 SDU, p<7e-05). We show that a published European GWAS of cognition predicted levels of executive function (average decrease in cognition of 0.06 SDU per unit increase in PRS, p<2e-04), but not diagnosis or symptoms. Specific loci identified in the SMI GWAS also showed association to multiple diagnoses, symptoms, and cognitive traits in Paisa. The most noteworthy result was for a locus on chromosome 7p22.3, associated in multiple SMI GWAS, that showed association in Paisa to increased risk of bipolar disorder, and to reduced complex cognition and social cognition. Our findings demonstrate wide portability from European GWAS to an admixed American sample, with associations to multiple transdiagnostic phenotypes.
Open AccessEffect of ancestry and shared genetic architecture of serious mental illness on symptoms and cognition in an admixed Latin American population
Most genome-wide association studies (GWAS) of serious mental illness (SMI) have been conducted for categorical diagnoses in samples of primarily European ancestry. The portability of findings to non-Europeans, and to SMI-related symptoms/dimensional traits remains uncertain. In a sample of 8,666 SMI cases and controls from the Paisa region of Colombia we show that a primarily European schizophrenia GWAS polygenic risk score (PRS) predicted all SMI diagnoses in this sample, as well as symptoms (assessed in cases only) and traits assessed agnostic to SMI diagnosis: a one SD unit (SDU) increase in this PRS was associated to decreased risk in cases of suicidal thoughts (OR=0.89, 95% confidence interval 0.84-0.94), depressed mood (OR=0.90, 95% confidence interval 0.85-0.95), and increased risk of delusions (OR=1.12, 95% confidence interval 1.06-1.18) and to decreased cognition (in cases and controls) across five distinct domains (average decrease in cognition of 0.065 SDU, p<7e-05). We show that a published European GWAS of cognition predicted levels of executive function (average decrease in cognition of 0.06 SDU per unit increase in PRS, p<2e-04), but not diagnosis or symptoms. Specific loci identified in the SMI GWAS also showed association to multiple diagnoses, symptoms, and cognitive traits in Paisa. The most noteworthy result was for a locus on chromosome 7p22.3, associated in multiple SMI GWAS, that showed association in Paisa to increased risk of bipolar disorder, and to reduced complex cognition and social cognition. Our findings demonstrate wide portability from European GWAS to an admixed American sample, with associations to multiple transdiagnostic phenotypes.
- Open Access
- 12 February 2026
- Published by JMIR Publications Inc. in JMIR Preprints
Generative Large Language Models for Psychiatric Concept Detection in Spanish-Language Electronic Health Records: Development and Evaluation Study (Preprint)
BACKGROUND The accurate detection of clinical symptoms in electronic health records (EHRs) is pivotal for advancing large-scale genetic and longitudinal studies in psychiatry. Free-text clinical notes are an essential source of symptom-level information, particularly in psychiatry; however, the automated extraction of symptoms from clinical text remains challenging. OBJECTIVE We aimed to evaluate the ability of open-source generative large language models (LLMs) to detect psychiatric symptom-level concepts from clinical text in Spanish and to develop a new fine-tuned, open-source LLM that can be shared without risk of exposing patient information. METHODS We tested 11 open-source generative LLMs for their ability to detect 109 psychiatric concepts from clinical text, using annotated EHR notes from a psychiatric hospital in Colombia. The LLMs were evaluated both “out-of-the-box” and after fine-tuning and compared against a traditional natural language processing (tNLP) method developed from the same data. To develop a fine-tuned LLM that can be shared without risk of exposing patient information, we created a novel synthetic dataset, leveraging real concept annotations while generating fully synthetic clinical text, and fine-tuned a top-performing LLM on these data. Fine-tuned model generalizability was assessed through cross-site evaluation using data from a second independent hospital in Colombia. RESULTS We show that while base LLM performance was poor to moderate (zero-shot: 0.2-0.6 macro-F1; few-shot: 0.2-0.74 macro-F1), it improved significantly after fine-tuning (0.75-0.86 macro-F1), with several fine-tuned LLMs outperforming the tNLP method. In total, 100 psychiatric concepts could be reliably detected (F1>0.8) using either a fine-tuned LLM or tNLP. Using the synthetic dataset, we developed Mistral-small-psych, a new LLM capable of detecting psychiatric concepts from Spanish clinical text with performance comparable to models trained on real EHR data (macro-F1=0.79) while remaining fully distributable. In cross-site evaluations, most LLMs generalized well, with small out-of-domain performance losses (macro-F1 decrease of 0.02-0.16). CONCLUSIONS This work provides the first large-scale, cross-institutional evaluation of state-of-the-art LLMs for psychiatric concept detection in Spanish EHRs, introduces a privacy-preserving pathway for building shareable clinical natural language processing models, and presents a new open model for extracting symptom-level psychiatric information from Spanish clinical text.
- 12 February 2026
- Published by JMIR Publications Inc. in JMIR Preprints
Open AccessGenerative Large Language Models for Psychiatric Concept Detection in Spanish-Language Electronic Health Records: Development and Evaluation Study (Preprint)
BACKGROUND The accurate detection of clinical symptoms in electronic health records (EHRs) is pivotal for advancing large-scale genetic and longitudinal studies in psychiatry. Free-text clinical notes are an essential source of symptom-level information, particularly in psychiatry; however, the automated extraction of symptoms from clinical text remains challenging. OBJECTIVE We aimed to evaluate the ability of open-source generative large language models (LLMs) to detect psychiatric symptom-level concepts from clinical text in Spanish and to develop a new fine-tuned, open-source LLM that can be shared without risk of exposing patient information. METHODS We tested 11 open-source generative LLMs for their ability to detect 109 psychiatric concepts from clinical text, using annotated EHR notes from a psychiatric hospital in Colombia. The LLMs were evaluated both “out-of-the-box” and after fine-tuning and compared against a traditional natural language processing (tNLP) method developed from the same data. To develop a fine-tuned LLM that can be shared without risk of exposing patient information, we created a novel synthetic dataset, leveraging real concept annotations while generating fully synthetic clinical text, and fine-tuned a top-performing LLM on these data. Fine-tuned model generalizability was assessed through cross-site evaluation using data from a second independent hospital in Colombia. RESULTS We show that while base LLM performance was poor to moderate (zero-shot: 0.2-0.6 macro-F1; few-shot: 0.2-0.74 macro-F1), it improved significantly after fine-tuning (0.75-0.86 macro-F1), with several fine-tuned LLMs outperforming the tNLP method. In total, 100 psychiatric concepts could be reliably detected (F1>0.8) using either a fine-tuned LLM or tNLP. Using the synthetic dataset, we developed Mistral-small-psych, a new LLM capable of detecting psychiatric concepts from Spanish clinical text with performance comparable to models trained on real EHR data (macro-F1=0.79) while remaining fully distributable. In cross-site evaluations, most LLMs generalized well, with small out-of-domain performance losses (macro-F1 decrease of 0.02-0.16). CONCLUSIONS This work provides the first large-scale, cross-institutional evaluation of state-of-the-art LLMs for psychiatric concept detection in Spanish EHRs, introduces a privacy-preserving pathway for building shareable clinical natural language processing models, and presents a new open model for extracting symptom-level psychiatric information from Spanish clinical text.
- Open Access
Assessing the feasibility of large-scale digital sensing for depression and anxiety: The Digital Mental Health Study
Data passively obtained from smartphones and wearables can provide nearly continuous objective information that enables quantification of states and traits across broad physiological, behavioral, and emotional domains impacted in mental health conditions, including depression and anxiety. Widespread application of such digital phenotyping could transform the assessment of depression and anxiety in research and clinical care, but the field has lacked well-powered longitudinal studies demonstrating the utility of this approach. This paper describes the design and implementation of the Digital Mental Health Study (DMHS), which collected up to 12 months of sensor data from iPhone and Apple Watch in more than 4,000 participants, a sample diverse by age, sex at birth, ethnicity, and depression symptom severity. To enable the use of these digital phenotypes to assay the complexity and heterogeneity of depression and anxiety, we designed a protocol of periodic self-report and interview-based scales optimized to assess elements of depression, anxiety, and perceived stress as broadly as possible while minimizing participant measurement burden. We report here the strategies used to recruit and enroll the DMHS sample, the process employed to develop study methods and protocols, and initial findings describing longitudinal symptom trajectories and demonstrating high participant engagement and adherence over a 12-month period.
Open AccessAssessing the feasibility of large-scale digital sensing for depression and anxiety: The Digital Mental Health Study
Data passively obtained from smartphones and wearables can provide nearly continuous objective information that enables quantification of states and traits across broad physiological, behavioral, and emotional domains impacted in mental health conditions, including depression and anxiety. Widespread application of such digital phenotyping could transform the assessment of depression and anxiety in research and clinical care, but the field has lacked well-powered longitudinal studies demonstrating the utility of this approach. This paper describes the design and implementation of the Digital Mental Health Study (DMHS), which collected up to 12 months of sensor data from iPhone and Apple Watch in more than 4,000 participants, a sample diverse by age, sex at birth, ethnicity, and depression symptom severity. To enable the use of these digital phenotypes to assay the complexity and heterogeneity of depression and anxiety, we designed a protocol of periodic self-report and interview-based scales optimized to assess elements of depression, anxiety, and perceived stress as broadly as possible while minimizing participant measurement burden. We report here the strategies used to recruit and enroll the DMHS sample, the process employed to develop study methods and protocols, and initial findings describing longitudinal symptom trajectories and demonstrating high participant engagement and adherence over a 12-month period.
- Open Access
Quantitative trait loci mapping of gene expression and chromatin accessibility in primary fibroblast reveals shared allelic effects between Latin American and European ancestries
Quantitative Trait Locus (QTL) analysis of molecular data has identified genetic variants associated with traits such as gene expression, and colocalization of these functional QTL with GWAS risk loci has offered insights into the genetic basis of human disease. We employed gene expression (RNA-seq) and chromatin accessibility (ATAC-seq) obtained from human primary fibroblasts to investigate quantitative trait loci (QTLs) in cohorts ascertained for bipolar disorder of European (n=150) and Latin American (n=96) ancestries. Leveraging data from three countries of origin (The Netherlands, Colombia, Costa Rica) within our cohort, we characterized differences among individuals at the SNP, gene, and accessible-chromatin levels to compute ancestry-specific expression (e)QTLs and chromatin-accessibility (ca)QTLs. Across ancestries, we observed R² ≥ 0.93 for eQTL effect sizes and R² ≥ 0.95 for caQTLs, indicating a high degree of concordance. Integrating chromatin data with expression and genotype information enabled precise fine-mapping of eQTLs, yielding 203 high-confidence (posterior probability > 90 %) regulatory pathways. In downstream analyses, transcriptome-wide (TWAS) and chromatin-wide (CWAS) association studies with brain- and skin-related GWAS identified 36 TWAS-significant genes and 77 CWAS-significant open chromatin regions. These findings underscore the shared genetic regulatory mechanisms across European and Latin American ancestries, while demonstrating that ancestry-specific reference panels enhance the accuracy of TWAS and CWAS in diverse populations.
Open AccessQuantitative trait loci mapping of gene expression and chromatin accessibility in primary fibroblast reveals shared allelic effects between Latin American and European ancestries
Quantitative Trait Locus (QTL) analysis of molecular data has identified genetic variants associated with traits such as gene expression, and colocalization of these functional QTL with GWAS risk loci has offered insights into the genetic basis of human disease. We employed gene expression (RNA-seq) and chromatin accessibility (ATAC-seq) obtained from human primary fibroblasts to investigate quantitative trait loci (QTLs) in cohorts ascertained for bipolar disorder of European (n=150) and Latin American (n=96) ancestries. Leveraging data from three countries of origin (The Netherlands, Colombia, Costa Rica) within our cohort, we characterized differences among individuals at the SNP, gene, and accessible-chromatin levels to compute ancestry-specific expression (e)QTLs and chromatin-accessibility (ca)QTLs. Across ancestries, we observed R² ≥ 0.93 for eQTL effect sizes and R² ≥ 0.95 for caQTLs, indicating a high degree of concordance. Integrating chromatin data with expression and genotype information enabled precise fine-mapping of eQTLs, yielding 203 high-confidence (posterior probability > 90 %) regulatory pathways. In downstream analyses, transcriptome-wide (TWAS) and chromatin-wide (CWAS) association studies with brain- and skin-related GWAS identified 36 TWAS-significant genes and 77 CWAS-significant open chromatin regions. These findings underscore the shared genetic regulatory mechanisms across European and Latin American ancestries, while demonstrating that ancestry-specific reference panels enhance the accuracy of TWAS and CWAS in diverse populations.