Messy HapSum figure

[Yexin-Zhang]

Hi Authors,

I'm seeking advice on optimizing STITCH parameters for a low-coverage WGS dataset with the following characteristics:

• ~500 genomes total (3.85X average)
  - 400 samples at 0.5-3X coverage
  - 100 samples at 3-15X coverage
• All samples should have derived from 13 founders

I've evaluated STITCH's performance using different K values (4, 10, and 26) on two 10Mb chunks. Initial K=26 was chosen based on 13 founders (2 haplotypes each).

Command used:

K_VALUES=(4 10 26)

$STITCH --bamlist $bamlist \
    --sampleNames_file $samplelist \
    --outputdir $OUTPUT_DIR \
    --chr $CHR \
    --regionStart $START \
    --regionEnd $END \
    --buffer 100000 \
    --posfile $WD/region_files/Stitch_positions.region$CHUNK.txt \
    --K $K \
    --nGen 160 \
    --niterations 40 \
    --shuffle_bin_radius 500 \
    --splitReadIterations NA \
    --method pseudoHaploid \
    --switchModelIteration 31 \
    --nCores $SLURM_NTASKS \
    --output_format bgvcf \
    --tempdir $TEMDIR \
    --expRate 1 \
    --plot_shuffle_haplotype_attempts TRUE \
    --plotHapSumDuringIterations FALSE \
    --output_haplotype_dosages TRUE
 

1. Region 1 HapSum plots for K=4, K=10, K=26:

2. Region 2 HapSum plots for K=4, K=10:

3. ShuffleHaplotype plots showed similar patterns across all runs:

Could you help interpret these convergence patterns in the HapSum plots?Are there specific parameters you'd recommend adjusting to improve performance? Thank you for your help.

Best,
Yexin