Multilayer data integration and cross-validation of human protein atlas datasets: a generative artificial intelligence driven roadmap to validate DNA target gene predictions of KRAB ZNF proteins

Our computational roadmap is based on the following hypothesis: The transcriptional expression of KRAB ZNF target genes is assumed to be inversely regulated in respect to KRAB ZNF expression. The Human Protein Atlas (HPA) offers a valuable resource for validating expression and localization of KRAB ZNF proteins in immunohistochemical (IHC) images across normal and cancer tissue sections by providing RNA seq data of tissues and cell lines. The KRAB domain initially identified in KOX1/ZNF10 (December 27th, 1988) [PMID2288909] is encoded in about 400 human C2H2 zinc finger genes [PMID37716846]. The KRAB domain of KOX1 fused to tetracycline repressor as TETR-KRAB showed in 1995 [PMID7891684] that transcriptional repression is an active process mediated by SMP1/TRIM28 utilized in CRISPRi applications [PMID 25448066]. Modern KRAB C2H2 zinc finger protein evolved during tetrapod evolution by insertion of an additional amino acid in the leucine zipper structure of ancestral KRAB domains (PRDM7) [PMID35162997]. Our team efforts outline a comprehensive roadmap for validating DNA target genes of human KRAB ZNF genes utilizing HPA data sets. Specificities and cross-reactivities of 42 antibodies (29 KRAB and 13nonKRAB) were assessed by high-density-peptide-array analysis. 57120 immunohistochemical (IHC) images of 20 cancer and 48 normal tissues using 269 antibodies/240 KRAB proteins, 11 non-KRAB antibodies/10 proteins were semi-quantitated by Definiens Developer XD 1.1 defining 5 subsets of KRAB ZNF expression categories and compared to RNA seq data available from the same tissue [PMID24309898]. Opposed to individual pairwise comparisons of RNA to protein levels, a combination of selected KRAB ZNF IHC subsets showed reduced expression in cancer tissues on RNA and protein levels. Data integration of published KRAB ZNF targets genes [PMID37730438, PMID28273063, PMID33326746] identified a gene set of about 100 genes that harbor from 100 to 833 KRAB ZNF binding sites, whereas 2256 genes had only one KRAB ZNF binding site. Our computational analyses reveal technical challenges and experimental pitfalls regarding strengths and weaknesses of IHC images related to limited linearities of IHC images and antibody specificities within KRAB and non-KRAB C2H2 multigene zinc finger families. Limitations of RNA seq data sets are mirrored such as splice variants, transcriptional and posttranscriptional processing, and posttranslational modifications. Besides obviously required technical QC improvements, our Polygenic Expression Network (PEN) Score selects individuals that share non-coding and coding single nucleotide polymorphisms (SNPs), see our animation (http://www.indymed.net/#interferon/ to streamline diversities in population genetics. Our AI toolbox approach presents combined multilayer efforts for improving DNA target gene predictions of KRAB zinc finger (ZNF) proteins in human tissues in phylo- and ontogenesis asking the HPA initiative for whole genome information.