INC module, Including Negative Controls

[abearab]

Tom's suggestion

• https://www.nature.com/articles/s41586-020-2099-x
• https://github.com/biohank/CRISPR_screen_analysis

Effect sizes for sgRNAs were calculated as previously described [17,39].

1. In brief, log2 fold enrichments of sgRNAs were first measured between two samples.
2. For any given phenotype, a median log2 fold enrichment of all negative control sgRNAs (non-targeting and safe sgRNAs) was measured and this median value was subtracted from log2 fold enrichments of all sgRNAs to account for systematic bias in screens.
3. Lastly, log2 fold enrichments of all sgRNAs were divided by the standard deviation of negative control sgRNAs to yield phenotype Z scores (pZ) of sgRNAs which we used as effect size of sgRNAs.
4. Effect size of a gene is the median value of all sgRNAs that target the gene. We used modified t-value scores as our phenotype scores for genes, which account for both consistency and strength of all sgRNA effects for given genes.

Our phenotype scores based on t-value scores were computed as:
$$\text{phenotype score (T-score)}=\frac{(U_{gene} − U_{ctrl})}{\sqrt[2]{\frac{S_{var}}{N_{exp}} + \frac{S_{var}}{N_{ctrl}}}}$$
where $U_{gene}$ is the median effect of all sgRNAs ($pZ$) for a given gene, $U_{ctrl}$ is the median effect of all negative control sgRNAs ($pZ$), and $S_{var}$ is $Var_{gene} × (N_{exp} − 1) + Var_{ctrl} × (N_{ctrl} − 1)$, where $Var_{gene}$ is the variance of sgRNA effects ($pZ$) for a given gene, $N_{exp}$ is the number of sgRNAs for a given gene and $N_{ctrl}$ is the average number of sgRNAs per gene in a given screen.

[abearab]

I tried this here https://github.com/abearab/DDRi/blob/main/integrate-screens.ipynb

[abearab]

This has been implemented successfully.

https://screenpro2.readthedocs.io/en/latest/PhenoScore.html#phenoscore-module