BiocAsia 2024

transmogR: A Bioconductor package to enable the easy
incorporation of variants into a reference transcriptome

Dr Stevie Pederson (They/Them)
Black Ochre Data Labs
The Kids Research Institute Australia / ANU

November 8, 2024

The PROPHECY Study

• Currently leading the transcriptomic analysis for the PROPHECY study
  • Preventing Renal, OPthalmic and Heart Events in CommunitY
• Study with ~1350 SA-based Aboriginal participants
  • Genomic Variants, DNA Methylation, Transcriptomics, Proteomics, Lipidomics, Metabolomics
  • Inflammatory Marker Panel, Retinal Scans, ECG, Urine Samples
  • \(>\) 1000 metadata categories
• Will be the largest existing indigenous dataset when collected

• All omics layers are taken from Whole Blood Samples

Type II Diabetes and Complications

For Aboriginal Australians, Type II Diabetes rates are extremely high with younger onset¹

Similar for Cardiovascular & Chronic Kidney Disease

• Both CKD and CVD can be complications from Type II Diabetes
• This increase in disease burden translates to increased mortality, reduced life expectancy, an increased reliance on the health system and lower quality of life

1. https://www.indigenoushpf.gov.au

Motivation

• Source of this risk remains unknown
  • After controlling for known external factors \(\implies\) still highly increased risk

• Is there a genetic component?
  • Regulatory regions, Coding Changes, Structural Variation, etc?
• Is there an epigenetic component?
  • DNA Methylation

If genetic risk \(\implies\) will we be able to see it using GRCh38?

Graph-based approaches for transcriptomics are still too immature

• Alex, Kim etc set up in consultation with community

Variation In Aboriginal Australians

• About 25% of genetic variation in Aboriginal Australians is unique (Easteal et al. 2020)
  
  

• Can we incorporate into transcriptomic analysis?
  
  
• STARconsensus (Kaminow et al. 2022) for genomic alignments
  • gene-level counts
  • transcript-level analysis?
  • Does it matter?

Consensus variants found in >50% of unrelated individuals within a population, using the 1000GP AFR population members and the complete 1000GP dataset (panhuman) in comparison to PROPHECY participants

transmogR

• Able to produce a variant-modified reference transcriptome
  • Using a transcriptomic reference \(\implies\) don’t need co-ordinates
  • STARconsensus returns co-ordinates relative to unmodified reference

• Can also create a variant-modified reference genome
  • Used by salmon as decoy sequences (Srivastava et al. 2020)
  • Not used for any coordinate-based information

• This allows for quantification using salmon (or kallisto)
  • Overdispersion estimates returned for transcript-level analysis (Baldoni et al. 2024)

• Enables creation of raw fasta files \(\implies\) still need to index as per normal
• Currently trying to find a way to create diploid, personalised references

Main Functions

• transmogrify(): Create variant-modified transcriptome sequences
  • Add optional tags to modified transcript names

• genomogrify(): Create variant-modified genome sequences
  • Add optional tags

• digestSalmon():
  • Returns a SummarizedExperiment with assays counts, scaledCounts, TPM, effectiveLength
  • Overdispersions & transcript lengths returned in rowData
  • Can handle divergent references, e.g. parY-masked + chrY-masked

Creating a New Reference

• Requires a VCF with variants
  • Or a GRanges object with ALT/REF
  • Only tested with SNPs + InDels <50bp
• Transcripts from GTF or exonsBy()
• Genome from fasta or BSgenome
• Takes about 25-30min
  • Need to improve parallelisation
  • Can be RAM intensive \(\implies\) generally run on a server/HPC
• Output Fasta files using Biostrings::writeXStringSet()

Does It Make Any Difference?

• Only tested using haploid transcriptomes
• Pilot Analysis using 6 participants
  • Included GRCh38 as worst case
  • Personalised GRCh38 as best-case
  • Also AFR, panhuman and PROPHECY consensus variant sets
• Ran STARconsensus + transmogR/salmon
  • Observations from STARconsensus (Kaminow et al. 2022) replicated
  • PROPHECY closest to personalised \(\implies\) GRCh38 least similar

Intent is to assess whether the PROPHECY consensus set shows improvement or even any impact

Technical Assessment

• Change is shown relative to personalised references
• Only a small number of fragments at the summary level
• PROPHECY consistently closest to personalised in
  1. Assigned Fragments
  2. Concordant Fragments
  3. Correct Orientation (ISF)

These are very small numbers but PROPHECY does appear to resemble the personalised

Technical Assessment

• Looking at reads moving between genes (GRCh38 \(\rightarrow\) PROPHECY)
• 1 Changed alignment \(\neq\) 1 changed count
• Far more divergent than summary statistics indicate
• Lots of change in HLA genes
  \(\implies\) stand-alone analysis
• Reads changing transcript within the same gene were mostly HLA

• Alignments were taken from bam files output by salmon
• 1 changed alignment \(\neq\) 1 changed count due to multi-mapping and model fitting

Cohort Level Analysis

• 93 participants selected for a CKD analysis by the El Osta group (Baker Institute)
• Performed a quick T2D vs No T2D analysis
• Ran analyses in parallel
  • GRCh38 vs PROPHECY-modified GRCh38
  • PAR-Y excluded for males, chrY excluded for females

• Compared log₂ ratios of scaled counts
  • Incorporates counts AND overdispersions
• Also transcript-level DTE results

• PAR-Y excluded to be careful wrangling PAR region variants
• DTE analysis was just a crude T2D vs No T2D
• May not have been well balances for this analysis as complications were the focus

Cohort Level Analysis

• log2-ratio of scaled counts within each participant
  • GRCh38 vs PROPHECY-modified
  • Showing average change (Mean logFC) vs variation (SD logFC)
  • Also baseline logCPM (from GRCh38)
• Some transcripts become far more variable (x \(\approx\) 0; y \(\uparrow\))
• Some become consistently higher/lower in signal estimates

Change in Transcript-Level Results

• DTE Ranks: T2D vs No T2D
• Most transcripts are also consistent
• Some do change ranks quite significantly
• Probably time to simulate data for confidence

Bonus Extra Self Promotion

• I personally struggle with the MEME-Suite & conda environment
  \(\implies\) developed motifTestR
  • May have re-invented the wheel
• Has a test for positional bias within sequences
  • Analogous to centrimo
• Also tests for enrichment within sequences
  • poisson, quasipoisson, hypergeometric, non-parametric
• Can cluster motifs and test as a cluster instead of individual motifs

Acknowledgements

Black Ochre Data Labs

• Alex Brown
• Jimmy Breen
• Alastair Ludington
• Yassine Souilmi
• Sam Godwin
• Liza Kretzschmar
• Sam Buckberry
• Holly Massacci
• Katharine Brown
• Adam Heterick
• Justine Clark
• Amanda Richards-Satour
• Bastien Llamas
• Mary Brushe
• Sarah Munns
• Rose Senesi
• Kaashifah Bruce

Baker Heart & Diabetes Institute

• Sam El-Osta
• Ishant Khurana
• Scott Maxwell
• Moshe Olshansky

National Centre for Indigenous Genomics

• Hardip Patel
• Azure Hermes

SAHMRI / Wardliparingga

• Natasha Howard
• Marlie Frank
• Odette Pearson
• Kim Morey

University of Sydney

• Jean Yang

ALIGN

• Johanna Barclay
• Annalee Stearne
• Louise Lyons

SAGC

• Paul Wang
• John Salamon
• Sen Wang
• Renee Smith

Victor Chang Cardiac Research Institute

• Jason Kovacic

References

Baldoni, Pedro L, Yunshun Chen, Soroor Hediyeh-Zadeh, Yang Liao, Xueyi Dong, Matthew E Ritchie, Wei Shi, and Gordon K Smyth. 2024. “Dividing Out Quantification Uncertainty Allows Efficient Assessment of Differential Transcript Expression with edgeR.” Nucleic Acids Res. 52 (3): e13.

Easteal, Simon, Ruth M Arkell, Renzo F Balboa, Shayne A Bellingham, Alex D Brown, Tom Calma, Matthew C Cook, et al. 2020. “Equitable Expanded Carrier Screening Needs Indigenous Clinical and Population Genomic Data.” Am. J. Hum. Genet. 107 (2): 175–82.

Kaminow, Benjamin, Sara Ballouz, Jesse Gillis, and Alexander Dobin. 2022. “Pan-Human Consensus Genome Significantly Improves the Accuracy of RNA-seq Analyses.” Genome Res. 32 (4): 738–49.

Srivastava, Avi, Laraib Malik, Hirak Sarkar, Mohsen Zakeri, Fatemeh Almodaresi, Charlotte Soneson, Michael I Love, Carl Kingsford, and Rob Patro. 2020. “Alignment and Mapping Methodology Influence Transcript Abundance Estimation.” Genome Biol. 21 (1): 239.