Develop tools that use genetic data to better classify complex human diseases

Project Information

ClassifyDiseases

Grant agreement ID: 101088901

DOI

10.3030/101088901

EC signature date 24 April 2023

Start date 1 August 2023

End date 31 July 2028

Funded under

• European Research Council (ERC)

Total cost

€ 1 975 333,00

EU contribution

€ 1 975 333,00

Coordinated by

AARHUS UNIVERSITET
Denmark

Periodic Reporting for period 1 - ClassifyDiseases (Develop tools that use genetic data to better classify complex human diseases)

Reporting period: 2023-08-01 to 2026-01-31

Summary of the context and overall objectives of the project

Approximately one in four people will develop a neurological or psychiatric condition. These brain disorders tend to be highly heterogeneous, meaning that two individuals with the same disease can have very different progressions or responses to medication.

To effectively treat a patient with a brain disorder, it is necessary to quickly and accurately identify their subtype. At present, patient subtypes are decided using only clinical information, and the process is highly suboptimal. For example, the available subtypes are often incomplete or poorly-defined, meaning that many patients are wrongly classified or can not be classified at all.

Previous work has shown that for many brain disorders, the classification of patients can be improved by incorporating genetic information. However, for this to become a reality, requires statistical tools that do not yet exist. My project will develop tools for classifying heterogeneous diseases based on clinical and genetic information, that will lead to better treatment of patients with brain
disorders.

I will ensure that my new tools are freely available and easy-to-use, so that other groups can construct classification models for many different types of diseases. Overall, this project has the potential to revolutionize how patients with heterogeneous diseases are treated, and accelerate the widespread use of precision medicine.

Work performed from the beginning of the project to the end of the period covered by the report and main results achieved so far

This project has developed a number of statistical tools for analyzing genome-wide association study (GWAS) data:

1 - LDAK-KVIK - the leading tool for mixed-model association analysis (MMAA).

LDAK-KVIK is a tool for testing genetic variants for association, that can be applied to both continuous and binary phenotypes, and to both homogeneous and heterogeneous datasets. Its two main features are accuracy and speed. Specifically, we showed that in when applied to 40 traits from UK Biobank, LDAK-KVIK was more powerful than BOLT-LMM (previously the most powerful MMAA tool), yet as fast as REGENIE (one of the fastest MMAA tools).

LDAK-KVIK was published in Nature Genetics in September 2025.. It is freely-available within our software package LDAK (www.dougspeed.com). In addition, we have made a bespoke website (www.ldak-kvik.com) that provides detailed documentation (e.g. video tutorials).

2 - GigaPRS - the leading tool for constructing polygenic risk scores (PRS).

In 2021, we released MegaPRS (PMID: 34234142). We have now developed an updated versions. The key new features are improved performance (GigaPRS is both faster and more computationally efficient than MegaPRS), and that MegaPRS2 can now accommodate both multiple traits and multiple ancestries. We have shown that when applied to results from a single GWAS, GigaPRS matches the performance of SBayesRC (which we consider the leading existing PRS tool), and when applied to multiple traits and/or ancestries, GigaPRS substantially outperforms PRS-CSx and JointPRS (the two leading tools for analyzing results from multiple GWAS).

GigaPRS is freely-available within our software package LDAK (www.dougspeed.com) and we hope to have a preprint by May 2026.

3 - LDAK-PBAT - the leading tool for pathway analysis.

LDAK-PBAT is a tool for testing gene pathways (sets of genes) for association with a phenotype. Its main advantage is that it uses a heritability-based framework, and as a result, is substantially more powerful than existing tools such as MAGMA and GSA-Mixer.

LDAK-PBAT is freely-available within our software package LDAK (www.dougspeed.com) and has been described within a preprint (Medrxiv, 2025).

4 - StratGWAS - a tool for performing subtype-aware genome-wide association studies (GWAS)

StratGWAS is designed for analyzing diseases with heterogeneous clinical features (e.g. where patients vary based on age of onset, diagnosis type or medication use). It first creates a transformed phenotype, where individuals are weighted based on their estimated genetic liability (i.e. a measure of how likely they were to develop the disease), then performs a GWAS using the transformed phenotype. We have shown that StratGWAS can increase power by up to 20% compared to standard GWAS.

StratGWAS is available as a standalone R package (https://github.com/JasperHof/StratGWAS) and has been described within a preprint (Medrxiv, ,2026).

Progress beyond the state of the art and expected potential impact (including the socio-economic impact and the wider societal implications of the project so far)

During the first three years of the project, we have created LDAK-KVIK and LDAK-PBAT, the most powerful tools for mixed-model association analysis (identifying genetic variants associated with a phenotype) and for pathway analysis (identifying sets of genes associated with a phenotype). We have also created GigaPRS, the more accurate and flexible tool for constructing polygenic scores (genetic prediction models). Lastly, we have developed StratGWAS, a tool for analyzing disease with heterogeneous clinical features (e.g. age of onset, diagnosis criteria and medication update).

We have made all of our new tools freely-available, either within our software package LDAK (www.dougspeed) or within a bespoke R package (https://github.com/JasperHof/StratGWAS).. We have taken special care to ensure our tools are easy to use. In particular, they are highly computationally efficient (most can be run on a laptop within an hour), and well-documented (e.g. the LDAK website provides extensive help guides, and users can ask questions at www.github.com/dougspeed/LDAK/issues).

We consider these tools both important in their own right, and for achieving the goals of this project, whose four main aims are as follows:

Aim 1 – Develop new SNP-based classification tools
Aim 2 – Construct better heritability models
Aim 3 – Develop methods to incorporate rare SNPs
Aim 4 – Develop GWAS tools that accommodate subtypes

Specifically, we believe we have provided a solution for Aim 4 (StratGWAS), believe we can address Aim 3 and 4 by generalizing LDAK-KVIK to accommodate rare variants, while we hope that GigaPRS and LDAK-PBAT can be instrumental in addressing Aims 1 and 2, respectively.

This project plans to create two new methods for disease classification based on genetic data. Type