kRISP-meR

My undergraduate thesis at BUET. kRISP-meR designs single-guide RNAs (sgRNAs) for CRISPR/Cas9 from the raw sequencing reads of the individual being edited, without a reference genome or a genome assembly.

Most guide-design tools check off-targets against a single reference genome. That is not possible for organisms without a good assembly, and it fails quietly when the individual differs from the reference: a guide can look unique in the reference while the same sequence is duplicated in the individual.

How it works:

• Reads are aligned to the user-supplied target sequence with Bowtie2 and the target is refined with Pilon, giving a personalised target from which candidate guides (20 nt protospacer plus NGG PAM) are enumerated
• All 23-mers in the reads are counted with Jellyfish, and the k-mer frequency spectrum is modelled as a mixture of Poisson distributions fitted with expectation-maximization, which yields the sequencing coverage and a prior over genomic copy number
• For each candidate, the expected number of cuts is computed over every k-mer within two mismatches, weighting the inferred copy number of each by its CFD cutting probability; guides are ranked by this value
• On-target efficiency is reported using the Rule Set 2 model of Doench et al.

Evaluation:

• Benchmarked against CRISPOR and GuideScan2 on human, mouse, and yeast: read-based scores closely track scores computed from the full reference genome, and the recommended guides carry far fewer severe off-targets
• Tested where individual and reference disagree (CHM13 versus GRCh38 in human, N22 versus Nipponbare in rice): kRISP-meR rejects guides that are single-copy in the reference but duplicated in the individual, and can evaluate targets that are absent from the reference altogether
• Benchmarking complete; manuscript being finalised