Andrea Ganna
@andganna.bsky.social
Associate Prof in health data science @HiLIFE_helsinki @FIMM_UH - MGH/Harvard - Playing with all kind of data - http://dsgelab.org
EU vice-president Virkkunen launches RAISE: resource for AI science in Europe. Not exactly clear what this entails. But i understand: more money, more networking.
November 3, 2025 at 2:36 PM
EU vice-president Virkkunen launches RAISE: resource for AI science in Europe. Not exactly clear what this entails. But i understand: more money, more networking.
Yoshua Bengio and the role of EU in AI development
November 3, 2025 at 1:57 PM
Yoshua Bengio and the role of EU in AI development
🎉 New preprint out!
"Removing genetic effects on plasma proteins enhances their utility as disease biomarkers"
We show that adjusting plasma proteins for genetic effects can make them stronger predictors of disease
👉 doi.org/10.1101/2025...
"Removing genetic effects on plasma proteins enhances their utility as disease biomarkers"
We show that adjusting plasma proteins for genetic effects can make them stronger predictors of disease
👉 doi.org/10.1101/2025...
October 16, 2025 at 9:16 PM
🎉 New preprint out!
"Removing genetic effects on plasma proteins enhances their utility as disease biomarkers"
We show that adjusting plasma proteins for genetic effects can make them stronger predictors of disease
👉 doi.org/10.1101/2025...
"Removing genetic effects on plasma proteins enhances their utility as disease biomarkers"
We show that adjusting plasma proteins for genetic effects can make them stronger predictors of disease
👉 doi.org/10.1101/2025...
💥 New preprint & our first RCT! 💥
Does an extremely high or low BMI polygenic score influence weight loss after a diet intervention?
GENEROOS is a 6-month randomized diet vs. control trial testing if genetic predisposition to higher BMI affects weight loss in overweight adults
Does an extremely high or low BMI polygenic score influence weight loss after a diet intervention?
GENEROOS is a 6-month randomized diet vs. control trial testing if genetic predisposition to higher BMI affects weight loss in overweight adults
October 8, 2025 at 8:01 PM
💥 New preprint & our first RCT! 💥
Does an extremely high or low BMI polygenic score influence weight loss after a diet intervention?
GENEROOS is a 6-month randomized diet vs. control trial testing if genetic predisposition to higher BMI affects weight loss in overweight adults
Does an extremely high or low BMI polygenic score influence weight loss after a diet intervention?
GENEROOS is a 6-month randomized diet vs. control trial testing if genetic predisposition to higher BMI affects weight loss in overweight adults
🧬💥 Do the genetics that make you develop a disease also help you survive it? Not much.
Our new study in Nature Genetics including 9 disease and 7 biobanks shows:
• Susceptibility variants ≠ survival
• PRSs for onset weak at predicting progression
• Lifespan PRS predicts survival better
Our new study in Nature Genetics including 9 disease and 7 biobanks shows:
• Susceptibility variants ≠ survival
• PRSs for onset weak at predicting progression
• Lifespan PRS predicts survival better
September 30, 2025 at 12:50 PM
🧬💥 Do the genetics that make you develop a disease also help you survive it? Not much.
Our new study in Nature Genetics including 9 disease and 7 biobanks shows:
• Susceptibility variants ≠ survival
• PRSs for onset weak at predicting progression
• Lifespan PRS predicts survival better
Our new study in Nature Genetics including 9 disease and 7 biobanks shows:
• Susceptibility variants ≠ survival
• PRSs for onset weak at predicting progression
• Lifespan PRS predicts survival better
Check out @tuomohartonen.bsky.social blog post on our study, out today in Nature Genetics, comparing electronic health record-based risk scores with genetic polygenic scores, and how they work better together!
tuomohartonen.substack.com/p/medical-re...
tuomohartonen.substack.com/p/medical-re...
August 27, 2025 at 5:47 PM
Check out @tuomohartonen.bsky.social blog post on our study, out today in Nature Genetics, comparing electronic health record-based risk scores with genetic polygenic scores, and how they work better together!
tuomohartonen.substack.com/p/medical-re...
tuomohartonen.substack.com/p/medical-re...
Impressive paper by EPIC. They built a autoregressive transformer on 150M individuals and 115 billion
medical events.
Including: diagnoses, medications, lab values, procedure
Cons: model is not shared (and unlikely will be)
arxiv.org/pdf/2508.12104
medical events.
Including: diagnoses, medications, lab values, procedure
Cons: model is not shared (and unlikely will be)
arxiv.org/pdf/2508.12104
August 22, 2025 at 5:57 AM
Impressive paper by EPIC. They built a autoregressive transformer on 150M individuals and 115 billion
medical events.
Including: diagnoses, medications, lab values, procedure
Cons: model is not shared (and unlikely will be)
arxiv.org/pdf/2508.12104
medical events.
Including: diagnoses, medications, lab values, procedure
Cons: model is not shared (and unlikely will be)
arxiv.org/pdf/2508.12104
Leena Peltonen School of Human Genetics in full-swing!
@gosiatrynka.bsky.social
@dgmacarthur.bsky.social
@bpasaniuc.bsky.social
@tuuliel.bsky.social
@hilarycmartin.bsky.social
@sashagusevposts.bsky.social
@zkutalik.bsky.social
@mashaals.bsky.social
@alemedinarivera.bsky.social
@gosiatrynka.bsky.social
@dgmacarthur.bsky.social
@bpasaniuc.bsky.social
@tuuliel.bsky.social
@hilarycmartin.bsky.social
@sashagusevposts.bsky.social
@zkutalik.bsky.social
@mashaals.bsky.social
@alemedinarivera.bsky.social
July 30, 2025 at 10:04 AM
Leena Peltonen School of Human Genetics in full-swing!
@gosiatrynka.bsky.social
@dgmacarthur.bsky.social
@bpasaniuc.bsky.social
@tuuliel.bsky.social
@hilarycmartin.bsky.social
@sashagusevposts.bsky.social
@zkutalik.bsky.social
@mashaals.bsky.social
@alemedinarivera.bsky.social
@gosiatrynka.bsky.social
@dgmacarthur.bsky.social
@bpasaniuc.bsky.social
@tuuliel.bsky.social
@hilarycmartin.bsky.social
@sashagusevposts.bsky.social
@zkutalik.bsky.social
@mashaals.bsky.social
@alemedinarivera.bsky.social
Personalized growth curves using results from longitudinal height GWAS of 80k japanese children
May 24, 2025 at 10:32 AM
Personalized growth curves using results from longitudinal height GWAS of 80k japanese children
May 24, 2025 at 8:42 AM
Behavioral and cognitive sciences got 30% funding boost despite very large cuts anywhere else. Interesting. Does anyone know why?
the other increase in funding is in area related to national security, which makes sense given the gov.
the other increase in funding is in area related to national security, which makes sense given the gov.
May 23, 2025 at 11:09 AM
Behavioral and cognitive sciences got 30% funding boost despite very large cuts anywhere else. Interesting. Does anyone know why?
the other increase in funding is in area related to national security, which makes sense given the gov.
the other increase in funding is in area related to national security, which makes sense given the gov.
We are looking for a post-doc/research fellow working on non-bullshit, reality-centered AI.
Meaning: unique nationwide EHR and -omics with the possibility to recontact people and prospectively test multimodal AI approaches.
jobs.helsinki.fi/job/Helsinki...
@FIMM_UH @FinnGen_FI
Meaning: unique nationwide EHR and -omics with the possibility to recontact people and prospectively test multimodal AI approaches.
jobs.helsinki.fi/job/Helsinki...
@FIMM_UH @FinnGen_FI
May 21, 2025 at 9:09 AM
We are looking for a post-doc/research fellow working on non-bullshit, reality-centered AI.
Meaning: unique nationwide EHR and -omics with the possibility to recontact people and prospectively test multimodal AI approaches.
jobs.helsinki.fi/job/Helsinki...
@FIMM_UH @FinnGen_FI
Meaning: unique nationwide EHR and -omics with the possibility to recontact people and prospectively test multimodal AI approaches.
jobs.helsinki.fi/job/Helsinki...
@FIMM_UH @FinnGen_FI
INTERVENE 2025 general meeting. What a group!
March 30, 2025 at 2:40 PM
INTERVENE 2025 general meeting. What a group!
Owning a cat but not a dog is associated with psychiatric disorders. Data from Danish National Birth Cohort.
February 27, 2025 at 1:04 PM
Owning a cat but not a dog is associated with psychiatric disorders. Data from Danish National Birth Cohort.
Here an example of the representation of semantic embedding for predictors of depression. Codes with similar meaning are close to each other, even if not present in all datasets.
It can be improved: for example, 'adult victim of abuse'
clustered near drug abuse concepts.
It can be improved: for example, 'adult victim of abuse'
clustered near drug abuse concepts.
February 5, 2025 at 10:04 AM
Here an example of the representation of semantic embedding for predictors of depression. Codes with similar meaning are close to each other, even if not present in all datasets.
It can be improved: for example, 'adult victim of abuse'
clustered near drug abuse concepts.
It can be improved: for example, 'adult victim of abuse'
clustered near drug abuse concepts.
We show that GRASP introduces significant inductive bias by positioning similar concepts nearby and unrelated concepts far apart, which can result in more efficient data utilization with fewer individuals.
February 5, 2025 at 10:04 AM
We show that GRASP introduces significant inductive bias by positioning similar concepts nearby and unrelated concepts far apart, which can result in more efficient data utilization with fewer individuals.
GRASP transfers well across datasets mapped to different data models accurately predicting disease risk from ICD-10-CM codes without direct mappings to OMOP.
So even if the model is trained and tested using different ontologies/vocabularies, GRASP still works.
So even if the model is trained and tested using different ontologies/vocabularies, GRASP still works.
February 5, 2025 at 10:04 AM
GRASP transfers well across datasets mapped to different data models accurately predicting disease risk from ICD-10-CM codes without direct mappings to OMOP.
So even if the model is trained and tested using different ontologies/vocabularies, GRASP still works.
So even if the model is trained and tested using different ontologies/vocabularies, GRASP still works.
Trained on the UK Biobank and evaluated in FinnGen and Mount Sinai, GRASP achieved an average ΔC-index that was 83% and 35% higher than language-unaware models, respectively. And outperformed XGboost.
February 5, 2025 at 10:04 AM
Trained on the UK Biobank and evaluated in FinnGen and Mount Sinai, GRASP achieved an average ΔC-index that was 83% and 35% higher than language-unaware models, respectively. And outperformed XGboost.
GRASP is a transformer that encode the semantic meaning of medical codes using LLMs (OpenAI). Thus it is less dependent on how or which codes are used.
We used GRASP to predict 21 health outcome using EHR data from >1 million people across three countries (UK, Finland, USA).
We used GRASP to predict 21 health outcome using EHR data from >1 million people across three countries (UK, Finland, USA).
February 5, 2025 at 10:04 AM
GRASP is a transformer that encode the semantic meaning of medical codes using LLMs (OpenAI). Thus it is less dependent on how or which codes are used.
We used GRASP to predict 21 health outcome using EHR data from >1 million people across three countries (UK, Finland, USA).
We used GRASP to predict 21 health outcome using EHR data from >1 million people across three countries (UK, Finland, USA).
Matthias visited our team from @hpi.bsky.social and had a great idea:
What if - to predict disease onset - we use the semantic meaning associated with structured EHR data, rather than the actual codes?
Would this make the model easier to transfer across countries/healthcare?
GRASP was born
What if - to predict disease onset - we use the semantic meaning associated with structured EHR data, rather than the actual codes?
Would this make the model easier to transfer across countries/healthcare?
GRASP was born
February 5, 2025 at 10:04 AM
Matthias visited our team from @hpi.bsky.social and had a great idea:
What if - to predict disease onset - we use the semantic meaning associated with structured EHR data, rather than the actual codes?
Would this make the model easier to transfer across countries/healthcare?
GRASP was born
What if - to predict disease onset - we use the semantic meaning associated with structured EHR data, rather than the actual codes?
Would this make the model easier to transfer across countries/healthcare?
GRASP was born
End of a two days visit to Angel Carracedo team at CiMUS and Galician Genome Project in Santiago de Compostela.
Reminds me that Europe has amazing local reality with excellence in science. Galicia has a unified EHR system covering 3M people since 20 years!
Reminds me that Europe has amazing local reality with excellence in science. Galicia has a unified EHR system covering 3M people since 20 years!
January 15, 2025 at 4:27 PM
End of a two days visit to Angel Carracedo team at CiMUS and Galician Genome Project in Santiago de Compostela.
Reminds me that Europe has amazing local reality with excellence in science. Galicia has a unified EHR system covering 3M people since 20 years!
Reminds me that Europe has amazing local reality with excellence in science. Galicia has a unified EHR system covering 3M people since 20 years!
In the UKBB proteomics pharma project, they are popping up an increasing number of non-big pharma partners. Some of them are interested in longevity and AI. Probably capturing the growing interest for proteomics and longevity.
January 10, 2025 at 11:07 AM
In the UKBB proteomics pharma project, they are popping up an increasing number of non-big pharma partners. Some of them are interested in longevity and AI. Probably capturing the growing interest for proteomics and longevity.
lovely! northern lights also here one hour north from helsinki
January 1, 2025 at 8:38 PM
lovely! northern lights also here one hour north from helsinki
🌟 Exciting news: The Leena Peltonen School of Human Genetics returns in 2025!
Back in 2015 I was a student. It was inspirational and helped me find my job at the Broad/Harvard
📅 July 27-31, 2025
📍 Wellcome Genome Campus, UK
📝 Apply by March 7 at www.lpshg.com
Back in 2015 I was a student. It was inspirational and helped me find my job at the Broad/Harvard
📅 July 27-31, 2025
📍 Wellcome Genome Campus, UK
📝 Apply by March 7 at www.lpshg.com
December 19, 2024 at 7:23 AM
🌟 Exciting news: The Leena Peltonen School of Human Genetics returns in 2025!
Back in 2015 I was a student. It was inspirational and helped me find my job at the Broad/Harvard
📅 July 27-31, 2025
📍 Wellcome Genome Campus, UK
📝 Apply by March 7 at www.lpshg.com
Back in 2015 I was a student. It was inspirational and helped me find my job at the Broad/Harvard
📅 July 27-31, 2025
📍 Wellcome Genome Campus, UK
📝 Apply by March 7 at www.lpshg.com