A 'writer-reader-eraser' post-translational modification regulatory system consisting of a large number of methyltransferases 8,9 , methyl-recognition domain-containing proteins 10 and putative demethylases 11 are expressed in different subcellular locations in humans, an indication that protein methylation is involved in processes other than epigenetic regulation.We prepared 82 Y2H bait strains spanning human R-methyltransferases (PRMT1-PRMT8) 8 , 16 SET domaincontaining K-methyltransferases (PKMTs) 9 , 9 members of the JMJD domain-containing protein family of protein demethylases and AOF2 (LSD1) 11 (Supplementary Table 1). In our current matrix screening protocol 4,12 , we perform four replicates, testing every set of baits individually against each of the ~13,000 prey contained in the matrix. Interacting prey are identified by their position in the matrix. To increase the sensitivity of the approach while also reducing the workload, we used a pooled strategy to test each protein pair substantially more than four times. Baits were pooled with all prey strains and then assayed for interaction in more than 100,000 separate spots ( Fig. 1a and Supplementary Fig. 1). Using Y2H-seq, we obtained 4-10 times the number of positive colonies obtained with the matrix approach. To reveal the prey identities, we collected all colonies and performed a 36-base parallel sequencing run. More than 20 million reads mapped perfectly to human RefSeq coding sequences (open reading frames, ORFs), corresponding to more than 500,000 unique 36-base reads (Supplementary Table 2). To rank the potentially interacting proteins for subsequent interaction retesting, we calculated a 'SeqScore' that incorporates the number of total mappings and the number of unique reads matching the ORF ( Supplementary Fig. 2). Notably, >99.7% of the RefSeq mappings matched to the 400 top-ranked genes, thus allowing the identification of potentially interacting ORFs with an extremely high signal-tobackground ratio (Supplementary Table 2).We performed four biological replicates and demonstrated in statistical pairwise comparisons that they result in very similar ranked prey orders (Supplementary Fig. 3). Top-ranked prey in at least two replicate screens were retested against all baits in a pairwise manner (Supplementary Fig. 4) and yielded 463 protein interactions (Supplementary Table 3). The success rate of the retest-that is, the probability that the prey is interactingdecreased with decreasing SeqScore (Fig. 1b).We also performed a matrix screen in quadruplicate with a subset of the protein methyltransferase (PMT) and protein demethylase (PDeM) baits for direct method comparison. With the matrix approach, we found 151 interactions (Supplementary to accelerate high-density interactome mapping, we developed a yeast two-hybrid interaction screening approach involving short-read second-generation sequencing (Y2h-seq) with improved sensitivity and a quantitative scoring readout allowing rapid interaction validation. We applied Y2h-seq to investigate enzymes in...