Refining GWAS

I refined the GWAS pipeline to include separate training and test sample sets, to:

  1. Examine multiple sets of markers based on various log10(pvalue) threshods (2-4.75), and select the best set based on the correlation between predicted trait value (e.g. HSI) v.s actual trait values. Here are figures showing results from the HSI markers.

16-hsi-logp2-manhattan

Genes with expression patterns also associated with performance in warming

I previously used WGCNA to identify genes with expression patterns that are associated with trait performance. Here are genes that were identified in that analysis AND in the above GWAS analysis to be associated with HSI. In bold are genes that are not only assigned to a HSI-correlated coexpressed gene module, but expression values are also correlated with HSI on their own.

Relationship with performance Chromosome Gene ID (NCBI) Uniprot SPID Protein
Increasing Expression NC_082382.1 LOC132453053 Q91453 Stonustoxin subunit beta (SNTX subunit beta) (DELTA-synanceitoxin-Sh1b) (DELTA-SYTX-Sh1b) (Trachynilysin subunit beta) (TLY subunit beta)
Increasing Expression NC_082383.1 LOC132456135 Q6UXZ4 Netrin receptor UNC5D (Protein unc-5 homolog 4) (Protein unc-5 homolog D)
Increasing Expression NC_082384.1 LOC132457513 Q91453 Stonustoxin subunit beta (SNTX subunit beta) (DELTA-synanceitoxin-Sh1b) (DELTA-SYTX-Sh1b) (Trachynilysin subunit beta) (TLY subunit beta)
Increasing Expression NC_082390.1 LOC132445594    
Increasing Expression NC_082390.1 LOC132452628 Q7RTR2 NLR family CARD domain-containing protein 3 (CARD15-like protein) (Caterpiller protein 16.2) (CLR16.2) (NACHT, LRR and CARD domains-containing protein 3) (Nucleotide-binding oligomerization domain protein 3)
Increasing Expression NC_082390.1 LOC132452644 Q9UM44 HERV-H LTR-associating protein 2 (Human endogenous retrovirus-H long terminal repeat-associating protein 2)
Decreasing Expression NC_082385.1 LOC132466560 Q5SVR0 TBC1 domain family member 9B
Decreasing Expression NC_082386.1 tmco1 Q6DGW9 Calcium load-activated calcium channel (CLAC channel) (GEL complex subunit TMCO1) (Transmembrane and coiled-coil domain-containing protein 1)
Decreasing Expression NC_082387.1 tmco1 Q6DGW9 Calcium load-activated calcium channel (CLAC channel) (GEL complex subunit TMCO1) (Transmembrane and coiled-coil domain-containing protein 1)

Identifying best set of markers for predicting trait value (here HSIU)

I pulled markers with log10(pvalue) = 2 - 4 at 0.25 increments, extracted genotype probabilities (imputed) for those n sites, then generated covariance matrices and performed PCA for each. I did this with ALL samples, not just training samples. This allowed me to assess agreements among predicted and actual trait values using modeled trait~PC1 scores relationships. The marker set that had the highest agreement in training and test sets were deemed the “best” set.

image image image

Here is the “final” list of HSI markers (n=39)

Chromosome Gene Number Markers in gene Protein Mean Log10(P) Uniqprot SPID
NC_082384.1 LOC132446484 3 Glutamate receptor ionotropic, delta-1 (GluD1) (GluR delta-1 subunit) 2.31106674 Q9ULK0
NC_082383.1 LOC132463408 2 Solute carrier family 25 member 36-A 2.396336651 Q6DG32
NC_082383.1 tmco1 2 Calcium load-activated calcium channel (CLAC channel) (GEL complex subunit TMCO1) (Transmembrane and coiled-coil domain-containing protein 1) 2.18904135 Q6DGW9
NC_082384.1 aatf 2 Protein AATF (Apoptosis-antagonizing transcription factor) 3.007019146 Q9QYW0
NC_082385.1 LOC132446484 2 Glutamate receptor ionotropic, delta-1 (GluD1) (GluR delta-1 subunit) 2.010495313 Q9ULK0
NC_082385.1 LOC132448018 2 Astrotactin-2 2.613209224 O75129
NC_082385.1 LOC132452589 2 Beta-1,4 N-acetylgalactosaminyltransferase 2 (EC 2.4.1.-) (Sd(a) beta-1,4-GalNAc transferase) (UDP-GalNAc:Neu5Aca2-3Galb-R b1,4-N-acetylgalactosaminyltransferase) 2.436358457 Q8NHY0
NC_082385.1 si:dkey-247m21.3 2 5-hydroxytryptamine receptor 4 (5-HT-4) (5-HT4) (Serotonin receptor 4) 2.239607094 O70528
NC_082382.1 LOC132456135 1 Netrin receptor UNC5D (Protein unc-5 homolog 4) (Protein unc-5 homolog D) 2.075278578 Q6UXZ4
NC_082382.1 LOC132457233 1 NLR family CARD domain-containing protein 3 (CARD15-like protein) (Caterpiller protein 16.2) (CLR16.2) (NACHT, LRR and CARD domains-containing protein 3) (Nucleotide-binding oligomerization domain protein 3) 3.223533315 Q7RTR2
NC_082382.1 LOC132461163 1   3.428210114  
NC_082382.1 LOC132461286 1   3.428210114  
NC_082382.1 LOC132463722 1   2.786885914  
NC_082382.1 LOC132463724 1   2.786885914  
NC_082382.1 LOC132463725 1   2.786885914  
NC_082382.1 acadm 1 Medium-chain specific acyl-CoA dehydrogenase, mitochondrial (MCAD) (EC 1.3.8.7) 2.786885914 Q3SZB4
NC_082382.1 cacna1hb 1 Voltage-dependent T-type calcium channel subunit alpha-1I (Voltage-gated calcium channel subunit alpha Cav3.3) (Ca(v)3.3) 2.136198402 Q9P0X4
NC_082383.1 LOC132463722 1   2.363614851  
NC_082383.1 LOC132463724 1   2.363614851  
NC_082383.1 LOC132463725 1   2.363614851  
NC_082383.1 LOC132463729 1   2.363614851  
NC_082383.1 LOC132466560 1 TBC1 domain family member 9B 2.238526242 Q5SVR0
NC_082383.1 bbs9 1 Protein PTHB1 (Bardet-Biedl syndrome 9 protein homolog) (Parathyroid hormone-responsive B1 gene protein homolog) 2.359376133 Q6AX60
NC_082383.1 trnap-agg_74 1   2.359376133  
NC_082383.1 trnap-agg_75 1   2.359376133  
NC_082383.1 uck2a 1 Uridine-cytidine kinase 2-A (UCK 2-A) (EC 2.7.1.48) (Cytidine monophosphokinase 2-A) (Uridine monophosphokinase 2-A) 2.198464599 Q7SYM0
NC_082384.1 LOC132469993 1 Desmoplakin (DP) (250/210 kDa paraneoplastic pemphigus antigen) 2.595272889 P15924
NC_082384.1 cpt2 1 Carnitine O-palmitoyltransferase 2, mitochondrial (EC 2.3.1.21) (Carnitine palmitoyltransferase II) (CPT II) 2.737512152 Q5U3U3
NC_082385.1 LOC132446534 1   2.010063808  
NC_082385.1 LOC132451248 1   2.445910261  

Performance of PC1 in predicting each trait in training samples (75%), test samples (25%), and all samples

In general, I’m a bit bummed to see that the correlations among predicted and actual trait values are not higher in our test sample sets - here’s a breakdown of how each gene set (based on min. log10(pvalue) did in predicting traits: image

Next step: Visualize genotype-dependent trait performance along with expression!

Written on December 5, 2024