[BulkRNASeq] Handling Technical Replicates

[J-81]

Description

Workflow should handle technical replicates appropriately.

Approaches

DESeq2 provides a collapseReplicates function that sums counts based on a factor to group samples by. The rationale has two major points:

1. Summing opposed to averaging is appropriate for maintaining expected Poisson distribution
2. DESeq2 is designed to normalize for library size differences. Summing technical replicates is akin to having a higher sequencing depth for a sample.

Implementation Suggested

Encode Technical Replicate Groups in the Runsheet

Encode technical replicates as a column in the runsheet simply using integers for each technical replicate group. Eventually, this technical replicate column should be automatically derived from ISA archive metadata; however, in the meantime, a workflow user should be able to supply a two column csv mapping sample name to technical replicate group which will be incorporated into the runsheet.

Use Technical Replicate Groups Column in Runsheet to for DESeq2 collapseReplicates

https://rdrr.io/bioc/DESeq2/man/collapseReplicates.html

Validation Plan

1. Validate reasonable approach results as follows:

Run the following approaches

• NF_RCP-F_1.0.3 (i.e. no technical replicate handling)
• collapseReplicates (summed tech. replicates)
• median replicates
• mean replicates
• filter to first replicate only (drop others)

Assessment Metrics:

• DGE results

2. Regression Test Criteria

• Core tests should run without change in outcomes (since core tests don't include any technical replicates)

[J-81]

Implementation Steps

• [ ] Runsheet generation now ingests optional technical replicate group table
• [ ] DESeq2 script updated to use tech. rep group data for four handling approaches

[J-81]

Additional considerations:

• How to handle technical replicates on multiple levels (e.g. multiple tissue cuts and multiple library preps for same biological sample)
• Group statistic and sample count
  • Perform after or before collapsing replicates?
  • Group stats related code does require reworking as currently written after collapsing replicates