Uneven sampling ranges of hash values are dependent on data input.

[jmonroynieto]

Extracting 5000 hashes and plotting their values per "strain"/genomicReads. After extracting the values of the hashes using the cap'nprot def, I noticed that regardless of seed used, one of my samples is getting a larger hash range for it's 5000 hashes. This goes against expectations since seed ranges should be comparable or at least not reliably bigger since the seed selection changes the hashing function and therefore produce new orderings of kmers with new aprox-random mappings which should not have range biases.

Can this bias be explained somehow? Could somone replicate to verify that the bias is not imposed by my implementation of data extraction or analyses?

This test involves 10 seed values [42 123456 987654 112233 445566 778899 135792 246813 314159 271828] and three SRR experiments representing three Cryptococcus gattii VGIII. strains SRR3707756:A - SRR3707833:B - SRR3707850:C

 cat read_1 read_2 | mash sketch \
            -k 21 \
            -s 5000 \
            -m 5 \
            -S "$seed" \
            -p "$MASH_THREADS" \
            -o "$outfile" \
            -

7 out of 10 seeds show the clear ordering median(A) > median(B) > median(C)

seed	A_gt_B	B_gt_C	consistent_order
42	TRUE	TRUE	TRUE
112233	TRUE	TRUE	TRUE
123456	TRUE	FALSE	FALSE
135792	TRUE	FALSE	FALSE
246813	TRUE	TRUE	TRUE
271828	TRUE	TRUE	TRUE
314159	TRUE	TRUE	TRUE
445566	TRUE	TRUE	TRUE
778899	TRUE	FALSE	FALSE
987654	TRUE	TRUE	TRUE