Making Polygenic Risk Scores More Accurate for Indonesians: A New Approach to Obesity Prediction

Photo by Google DeepMind on Unsplash

“Human beings are poor examiners, subject to superstition, bias, prejudice, and a PROFOUND tendency to see what they want to see rather than what is really there” — M. Scott Peck

Obesity is a growing health concern in Indonesia, with prevalence rates rising from 10.3% in 2007 to 23.4% in 2023. While environmental factors such as diet and physical activity play a crucial role, genetics also significantly influences obesity risk. Scientists estimate genetic susceptibility using Polygenic Risk Scores (PRS) is a statistical model that aggregates the effects of multiple genetic variants associated with obesity.

Increase in obesity prevalence among the population aged over 18 years from 2007 to 2023

However, a key limitation of current PRS models is that they are predominantly developed using European ancestry data, making them less effective for individuals from non-European populations, including Indonesians. Given that 91% of genome-wide association studies (GWAS) focus on European cohorts, PRS models often fail to generalize across genetically diverse populations. This study aims to address this issue by developing an ancestry-adjusted PRS for obesity, improving prediction accuracy for the Indonesian population.

Current GWAS focus heavily on European which limits applicability for Indonesians

What is a Polygenic Risk Score (PRS)?

A Polygenic Risk Score (PRS) is a number that estimates a person’s genetic risk for a disease or trait based on their DNA.

Every person’s DNA contains millions of genetic variations (SNPs), and some of these are linked to health conditions like obesity, diabetes, or heart disease. Instead of looking at just one gene, a PRS adds up the effects of many small genetic variations across the entire genome to predict a person’s likelihood of developing a condition.

Polygenic Risk Score Calculations

Think of it like a credit score for health:

• Just like a credit score predicts financial risk based on different factors (income, debt, payment history),
• A PRS predicts disease risk based on genetic factors inherited from your parents.

However, PRS is not a diagnosis — it simply indicates whether someone has a higher or lower genetic tendency for a disease compared to others.

Because PRS is usually developed using genetic data from European populations, it may not work as well for people from other backgrounds, like Indonesians. That’s why adjusting PRS for ancestry is important!

Why Do We Need an Ancestry-Adjusted PRS?

PRS models work by summing the effects of single nucleotide polymorphisms (SNPs) that have been linked to a particular trait. For obesity, key SNPs in genes like FTO, MC4R, and TMEM18 have been identified as major contributors. However, because allele frequencies and linkage disequilibrium (LD) patterns vary between populations, a PRS trained on Europeans might not accurately predict obesity risk in Indonesians.

A well-known challenge in genetic prediction models is population structure, which refers to genetic differences between populations due to historical migration patterns and evolutionary pressures. If population structure isn’t accounted for, PRS can be biased, overestimating or underestimating genetic risk in different ethnic groups. To correct for this, we implemented an ancestry-adjustment step using Principal Component Analysis (PCA) to make PRS more applicable to Indonesians.

How We Built an Ancestry-Adjusted PRS for Indonesians

To ensure a robust ancestry-adjusted model, we used:

Data Processing Steps

• Reference Panel: Global genetic data from the 1000 Genomes Project
• Indonesian Dataset: Genomic data from 2,800 Indonesian individuals

Step-by-Step Methodology

Ancestry-Adjustment Analysis Steps

1. Raw PRS Calculation

• We applied a PRS model trained on European GWAS data, which aggregates obesity-associated SNPs into a single risk score for each individual.

1. Principal Component Analysis (PCA) for Ancestry Adjustment

• Using the 1000 Genomes dataset, we extracted four principal components (PCs) to capture population structure. These PCs represent ancestry-related variation and are commonly used in genetic studies to correct for confounding due to population differences.

1. Linear Regression for PRS Adjustment

• We trained a linear regression model that predicts PRS as a function of ancestry PCs. This allowed us to estimate how much of an individual’s PRS score was driven by ancestry-related differences rather than true genetic risk.

1. Ancestry-Adjusted PRS Calculation

• We subtracted the ancestry-predicted PRS component from the raw PRS to obtain an adjusted score that better reflects true obesity risk for Indonesians.

Key Findings

Plot among races Before (right) vs. After (left) Ancestry adjustment

Before Ancestry Adjustment

• PRS values varied significantly across populations, with Indonesians showing systematic differences compared to Europeans, suggesting ancestry-related bias.

After Ancestry Adjustment

• PRS distributions became more homogeneous between populations, indicating reduced bias.
• Accuracy improved, as shown by:
• Mean Squared Error (MSE) reduction from 0.0148 to 0.0054
• PRS standard deviation drop from 0.430 to 0.259
• This demonstrates that our ancestry-adjusted PRS provides a more reliable and fair genetic risk score for obesity in Indonesians.

Implications for Personalized Medicine

By developing an ancestry-aware approach to PRS, this study highlights the need for more inclusive genetic risk prediction models. The implications extend beyond obesity:

✅ More accurate risk prediction for Indonesian patients ✅ Personalized healthcare interventions based on genetic risk ✅ Potential for expanding to other diseases like diabetes and cardiovascular conditions ✅ A roadmap for improving PRS models in other underrepresented populations

As precision medicine becomes increasingly important, ensuring that genetic models are tailored to diverse populations is critical for equitable healthcare.

Conclusion

Our study presents a novel method for improving genetic risk prediction in Indonesians by adjusting PRS for ancestry. By incorporating population structure corrections, we move closer to developing truly inclusive genomic tools that benefit individuals worldwide. Future research should focus on refining these models further, integrating multi-ancestry GWAS data, and expanding their applications to other complex diseases.