Isoform-level transcriptome-wide association uncovers genetic risk mechanisms for neuropsychiatric disorders in the human brain.

"the developmental brain rna-seq and genotype dataset from walker et al is available at dbgap with accession number phs001900 (ref 22 accesible at https://www ncbi nlm nih gov/projects/gap/cgi-bin/study cgi?; gwas summary statistics and accession numbers to genotype and rna-seq data are provided in supplementary table 10 isotwas models for 48 tissues from gtex are available at https://zenodo org/record/8047940 (ref 86 ) adult brain cortex from psychencode and amp-ad are available at https://zenodo org/record/8048198 (ref 87 ) and developmental brain cortex from walker et al are available at https://zenodo org/record/8048137 (ref 88 ). gwas summary statistics and accession numbers to genotype and rna-seq data are provided in supplementary table 10 isotwas models for 48 tissues from gtex are available at https://zenodo org/record/8047940 (ref 86 ) adult brain cortex from psychencode and amp-ad are available at https://zenodo org/record/8048198 (ref 87 ) and developmental brain cortex from walker et al are available at https://zenodo org/record/8048137 (ref 88 ). for isotwas multivariate elastic net demonstrated the greatest cv prediction of isoform expression across most simulation settings (fig 2b extended data fig 2a and supplementary data 1 ).; for total gene expression prediction the optimal isotwas models in sum outperformed the optimal twas model particularly at sparser isoqtl architectures with median absolute increase in adjusted r 2 of 0 6-3 5% (fig 2c extended data fig 2b and supplementary data 2 ).; in simulations isotwas prediction of gene expression also increases as the proportion of shared non-zero effect snps across isoforms decreases (fig 2bc extended data fig 2b and supplementary data 2 ).; as genes differ in the number and expression patterns of their constituent isoforms gene length snp density quantification accuracy and other relevant factors we characterized their impact on isotwas performance ( methods supplementary note extended data fig 6 and supplementary data 3 and 4 ).; first the fpr is controlled at 0 05 for isoform-level mapping using acat (extended data fig 7a and supplementary data 5 ).; scenario 1 showed clear increases in power for twas over isotwas but this advantage decreased with increased causal proportion of isoqtls and proportion of shared isoqtls (fig 4a and supplementary data 6 ).; for scenarios2 and 3 as effects on the trait varied across isoforms of the same gene (fig 4bc and supplementary data 7 and 8 ) isotwas showed clear increases in power over twas across most scenarios and causal effect architectures and particularly in settings with one effect isoform or two divergent effect isoforms.; finally we assessed the performance of probabilistic fine mapping in identifying the true effect isoform in our simulation framework of genes with 5 or 10 isoforms ( methods extended data fig 7c and supplementary data 9 ).; we detected more trait-associated genes with isotwas compared with twas across adult (2595 versus1589 genes) and developmental (4062 versus890 genes) reference panels respectively (extended data fig 8b and supplementary data 10 - 13 ).; we illustrate several examples of isotwas-prioritized isoforms all in the adult cortex for genes with limited or distinct expression qtls (fig 6 extended data fig 10b and supplementary data 14 ) with exon/intron structure shown in supplementary figs1 - 4 .; supplementary data 1 predictive performance comparison of isotwas multivariate methods in simulated data across a variety of genetic architecture settings.; supplementary data 2 predictive performance comparison of isotwas and twas gene expression prediction in simulated data across a variety of genetic architecture settings.; supplementary data 3 isoform expression prediction metrics across a variety of factors using 48 gtex datasets.; supplementary data 4 gene expression prediction metrics across a variety of factors using 48 gtex datasets.; supplementary data 5 false positive rates using isotwas and twas to detect a gene-trait association at p < 0 05 across a variety of genetic architecture parameters.; supplementary data 6 power to detect trait association at p < 2 5 x 10 -6 across1000 simulations each for 22 genes using twas and isotwas across various simulations.; supplementary data 7 power to detect trait association at p < 2 5 x 10 -6 across1000 simulations each for 22 genes using twas and isotwas (acat) across various simulations.; supplementary data 8 power to detect trait association at p < 2 5 x 10 -6 across1000 simulations each for 22 genes using twas and isotwas (acat) across various simulations.; supplementary data 9 sensitivity and mean set size of 90% credible sets determined by focus in simulated data across a variety of genetic architecture parameters.; supplementary data 10 raw twas results across15 neuropsychiatric traits using adult brain cortex expression models.; supplementary data 11 raw isotwas results across15 neuropsychiatric traits using adult brain cortex expression models.; supplementary data 12 raw twas results across15 neuropsychiatric traits using developmental brain cortex expression models.; supplementary data 13 raw isotwas results across15 neuropsychiatric traits using developmental brain cortex expression models.; supplementary data 14 gwas and nominal eqtl and isoqtl summary statistics corresponding to isotwas isoform-trait association examples shown in fig 6 and extended data fig 10b."

"code availability isotwas is available as an r package at https://github com/bhattacharya-a-bt/isotwas (ref 47 ).; sample scripts for analyses are available at https://github com/bhattacharya-a-bt/isotwas_manu_scripts (ref 114 )."

"Competing interests The authors declare no competing interests."

"We thank Kangcheng Hou, Tommer Schwarz, Vidhya Venkateswaran, Pan Zhang, Leanna Hernandez, Nathan LaPierre, Harold Pimentel, Mike Love and Achal Patel for engaging discussion during the research process. We thank Kanishka Patel for her aesthetic advice for figures. We thank the Psychiatric Genomics Consortium and Complex Trait Genomics Lab for their publicly available GWAS summary statistics. This work was supported by National Institutes of Health awards R01 HG009120, R01 MH115676, R01 CA251555, R01 AI153827, R01 HG006399, R01 CA244670 and U01 HG011715 (B.P.), as well as SFARI Bridge to Independence Award, NIMH R01-MH121521, NIMH R01-MH123922 and NICHD-P50-HD103557 (M.J.G.). The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript."