Skip to content

DecodeME/Multisite Pain

I applied Bivariate MiXeR2 to summary statistics from DecodeME3 and Johnston et al.'s GWAS of multisite pain4 in an attempt to quantify their joint genetic architecture.

In the standard workflow suggested by authors of bivariate MiXeR, 20 separate bivariate MiXeR models are trained, each on a different random subset of a reference panel of SNPs. This serves as a form of bootstrapping. Each of these models are then evaluated on the full reference panel. Unfortunately, MiXeR is memory-hungry, requiring something like 32GB for this evaluation step. To reduce memory requirements, I evaluated on random subsets of the reference panel instead of the full reference panel. In the future, I may rent a cloud machine to do full evaluation on the entire reference panel.

The key output parameters from my bivariate MiXeR runs are shown below:

Parameter Value
trait1 DecodeME
trait2 Multsite_pain
dice (mean) 0.5353
dice (std) 0.06681
pi1 (mean) 0.0003357
pi1 (std) 0.000224
pi2 (mean) 0.002353
pi2 (std) 0.0002129
pi12 (mean) 0.001547
pi12 (std) 0.0001974
nc1@p9 (mean) 1071
nc1@p9 (std) 714.5
nc2@p9 (mean) 7503
nc2@p9 (std) 679
nc12@p9 (mean) 4932
nc12@p9 (std) 629.4
rho_zero (mean) 0.02812
rho_zero (std) 0.002059
rho_beta (mean) 0.8738
rho_beta (std) 0.1033
rg (mean) 0.4923
rg (std) 0.01067
fraction_concordant_within_shared (mean) 0.857
fraction_concordant_within_shared (std) 0.078
best_vs_min_AIC -0.7013
best_vs_min_BIC -10.27
best_vs_max_AIC -0.7009
best_vs_max_BIC -10.26

The most important observation from these output parameters is that best_vs_min_AIC and best_vs_max_AIC are both slightly negative. According to the README provided by the authors this indicates that there is not strong statistical support for the full bivariate MiXeR model over simpler statistical models. I assume that this finding is a consequence of noise in the DecodeME summary statistics due to the relatively small sample size of the DecodeME study1.

This weak statistical support for application of bivariate MiXeR means that findings from bivariate MiXeR for this pair of traits need to be taken with a grain of salt. We can still analyze bivariate MiXeR's out, but this analysis should be regarded as exploratory, not definitive.

With this caveat, the main finding of bivariate MiXeR applied to ME/CFS and Multisite pain is that the two traits share a large polygenic overlap. This result is illustrated in the standard MiXeR plot below:

me_pain_bivariate_mixer

The conclusion from the Venn diagram is rather striking: as far as bivariate MiXeR can tell, the biological processes that cause ME/CFS are mostly a subset of the biological processes that cause chronic, multisite pain. Only a relatively small number of genetic variants truly distinguish ME/CFS.

  1. MiXeR computes AIC as part of the model fitting procedure, so AIC is not affected by my decision to evaluate on random subsets of the reference panel instead of the full reference panel. ↩

  2. Oleksandr Frei, Dominic Holland, Olav B Smeland, Alexey A Shadrin, Chun Chieh Fan, Steffen Maeland, Kevin S O’Connell, Yunpeng Wang, Srdjan Djurovic, Wesley K Thompson, and others. Bivariate causal mixture model quantifies polygenic overlap between complex traits beyond genetic correlation. Nature Communications, 10(1):2417, 2019. URL: https://www.nature.com/articles/s41467-019-10310-0. ↩

  3. Genetics Delivery Team, Thibaud Boutin, Andrew D Bretherick, Joshua J Dibble, Esther Ewaoluwagbemiga, Emma Northwood, Gemma L Samms, Veronique Vitart, Project, Cohort Delivery Team, Øyvind Almelid, and others. Initial findings from the DecodeME genome-wide association study of myalgic encephalomyelitis/chronic fatigue syndrome. medRxiv, pages 2025–08, 2025. URL: https://www.medrxiv.org/content/10.1101/2025.08.06.25333109v1. ↩

  4. Keira JA Johnston, Mark J Adams, Barbara I Nicholl, Joey Ward, Rona J Strawbridge, Amy Ferguson, Andrew M McIntosh, Mark ES Bailey, and Daniel J Smith. Genome-wide association study of multisite chronic pain in UK Biobank. PLoS Genetics, 15(6):e1008164, 2019. URL: https://journals.plos.org/plosgenetics/article?id=10.1371/journal.pgen.1008164. ↩