TITANS (TrIo-based Transcriptome-wide AssociatioN Study) is a statistical framework to conduct TWAS in proband-parent trios. TITANS quantifies the transmission disequilibrium of genetically regulated gene expression from parents to probands using a pseudosibling simulation approach and conditional logistic regression.

The software is developed and tested in Linux and Mac OS environments. The statistical computing software R (>=3.5.1) and the following R packages are required for association tests:

• data.table (>=1.11.8)
• dplyr (>=0.8.3)
• optparse (>=1.6.0)
• reshape2 (>=1.4.3)
• survival (>=2.44-1.1)
• tidyverse (>=1.2.1)

TITANS is a statistical tool for analysing phased trio genotyped data. We recommand performing phasing and imputation using Michigan Imputation Server. You can use bcftools and the following codes to perform post-imputation quality control

$ bcftools filter -i 'INFO/MAF>0.01 && INFO/R2>0.8' $INPUT -o $OUTPUT

where $INPUT is the vcf.gz file processed by Michigan Imputation Server. The bcftools will generate a QCed vcf file $OUTPUT.

Note: For vcf.gz or vcf files not from Michigan Imputation Server, they need fields INFO/MAF and INFO/R2 to perform quality control.

$ git clone https://github.com/qlu-lab/TITANS.git
$ cd TITANS

$ wget ftp://ftp.biostat.wisc.edu/pub/lu_group/Projects/TITANS/Software/GeneExpressionModels.zip
$ unzip GeneExpressionModels.zip

After upzipping, there will be directories containing gene expression imputation models from 12 brain tissues from Genotype-Tissue Expression (GTEx) and CommonMind consortium (CMC).

Each model is stored in separate file under ./$TISSUE/pred$CHR/pred.$GENE.txt, for example, the gene expression imputation model of POU3F2 on chromosome 6 in tissue Brain_Hippocampus is stored under ./Brain_Hippocampus/pred6/pred.POU3F2.txt. Each gene expression imputation model contains 6 fields:

We provide a walkthrough using toy data. First, create a dirctory for the output

$ mkdir Results ## under TITANS/

and use the toy data under Example to conduct a trio-based TWAS on gene toy on chromosome 1 in toy_tissue, the gene expression imputation model for this toy gene is under Example/toy.pred.txt, and the QCed vcf file and fam file are under ./Example/toy.vcf and ./Example/toy.fam respectively.

$ Rscript TITANS.assoc.R \
  --tissue toy_tissue \
  --chr 1 \
  --gene toy \
  --fam ./Example/toyfam \
  --pred ./Example/toy.pred.txt \
  --vcf ./Example/toy.vcf \
  --out ./Results/toy.txt
  --qc F \
  --matching 3sib

TITANS will write out a result table ./Results/toy.txt

Interpretation: This trio-based TWAS was based on gene toy on chromosome 1, with 80 SNPs in the gene expression imputation model. After post-imputation QC and checking for variant consistency between gene expression model and genotyping data, 72 SNPs remain in theanalysis. The disease effect size of gene toy estimated by three pseudo sibling matching and conditional logistic regression is 0.01522, while the direction of Beta indicates that the higher expression raises the disease risk. The P-value is 0.60149, indicating that the gene is not significantly associated with the disease. In other words, toy is not considered associating with the disease in this tissue.

The TITANS.assoc.R outputs the trio-based transcriptome-wide association test result one gene in one tissue at a time. TITANS.assoc.R takes the following inputs

$ Rscript TITANS.assoc.R \
    --tissue $TISSUE \
    --chr $CHR \
    --gene $GENE \
    --fam $FAMPATH \
    --pred $PRED \
    --vcf $VCF \
    --out $OUTPUT \
    --qc $QC \
    --matching $MATCHING

where the inputs are

We provided three matching methods for analyzing the trios. Matching method 3sib generates three pseudo siblings and estimates the effect size using conditional logistic regression, as illustrated in the workflow. Matching method 1sib generates 1 pseudo sibling using untransmitted parental alleles and estimates the effect size using logistic regression. Finally, 2parent estimates the effect size using the gene expressions from parents and conditional logistic regression.

The final result has the following fields:

You can use TITANS.organize.R and Tables/gene.coordinate.txt for adding gene coordinate information or mapping the intron clusters to genes. Tables/gene.coordinate.txt containing gene name, splicing ID (genes in CMC splicing will have its unique slicing ID, otherwise splicing ID is set as gene name), chromosomal location, and gene coordinate information based on hg19. Use the following code to organize your TITANS results.

$ Rscript
TITANS.organize.R \
    --input $INPUT \
    --infoTable $INFOTABLE \
    --out $OUTPUT \

where the inputs are

Note: If users provide wrong chromosomal and gene information, the result table will still be generated, but the gene coordinate will not be available.

If you use TITANS, please cite

Huang K, et al. (2021). Transcriptome-wide transmission disequilibrium analysis identifies novel risk genes for autism spectrum disorder. PLOS Genetics, 17(2): e1009309.

Kunling Huang (University of Wisconsin-Madison, Department of Statistics)

Yuchang Wu (University of Wisconsin-Madison, Department of Biostatistics and Medical Informatics)

Yupei Lin (University of Wisconsin-Madison, Department of Computer Sciences)

Qiongshi Lu (University of Wisconsin-Madison, Department of Biostatistics and Medical Informatics)

All rights reserved for Lu-Laboratory.

The imputation models and part of the codes are modified from the precomputed gene expression imputation model UTMOST and FUSION, we thank the authors for sharing their gene expression imputation model and codes.