Global genetic interaction network of a human cell maps conserved principles and informs functional interpretation of gene co-essentiality profiles

Fig. S1. Identifying essential genes and measuring single mutant fitness in HAP1 cells, Related to STAR Methods.

(A) Experimental setup for performing parallel, genome-scale pooled CRISPR screens in parental HAP1 cells with the TKOv3 library to measure single mutant fitness (LFC = log2 Fold Change).
(B) Workflow for generating a random forest model to identify 1,524 essential genes based on HAP1 screen data.
(C) Box plots showing the distribution of single mutant fitness effects for HAP1 genes grouped into the indicated fitness categories. DepMap60 includes genes that were classified as essential (CERES < -0.5) in at least 60% of cell lines tested in the DepMap 20Q2 dataset. The number of genes in each category is indicated in brackets. Individual perturbation of ~22% (3,941/17,724) of library genes significantly altered HAP1 cell fitness.
(D) Scatter plot of single mutant fitness effects (mean Log Fold Changes) derived from parental HAP1 screens in rich (n=18) or minimal (n=21) media. Single mutant fitness measurements derived from both growth conditions were highly correlated, with only a few genes showing significant condition-specific growth phenotypes (~0.2%, 38 genes).

Fig. S2. Characterization of HAP1 essential genes, Related to STAR Methods.

(A) Venn diagram showing the overlap between HAP1 essential genes and DepMap60 genes. Approximately, 81% (1231/1524) of HAP1 essential genes are also required for viability of most DepMap cancer cell lines. (i) Bar plot reporting the number of HAP1-specific essential genes (293 genes) that show DepMap essential phenotypes (mean CERES < -0.5) in the indicated fraction of DepMap cell lines. Genes not targeted by DepMap CRISPR libraries are indicated. (ii) Bar plot reporting the number of DepMap60-specific essential genes (383 genes) exhibiting fitness effects in HAP1 cells. Approximately, 19% (293/1524) of library genes were specifically essential in HAP1 cells but their perturbation often impacts growth of multiple cancer cell lines.
(B) Distributions showing the number of HAP1 essential genes (i) and non-essential genes (ii) that exhibited a similar phenotype in DepMap cell lines (number of DepMap cell lines for which CERES < -0.5).
(C) (i) Distribution illustrating the percentage of essential genes (CERES < -0.5) per DepMap cell line. Dotted line indicates the average number of essential genes per cancer cell line in DepMap. (ii) Distribution showing the percent of DepMap cell line-specific essential genes per cell line. Dotted vertical line indicates the average percentage of cell line-specific essential genes identified in a given DepMap cancer cell line. The percentage of HAP1-specific essentials is indicated. (iii) Distribution illustrates the percentage of DepMap60 genes that are nonessential (CERES > -0.5) in a particular cancer cell line. The average percentage of DepMap60 genes that are non-essential for viability in a specific cancer cell line and the percentage of DepMap60 genes nonessential in HAP1 cells are indicated (dotted lines).
(D) Functional enrichment among HAP1 essential genes. Reactome Pathway terms, ranging in size from 10 to 300 genes, statistically enriched (hypergeometric test, Benjamini-Hochberg-corrected FDR < 0.2) among HAP1 essential genes are summarized according to the functional descriptions shown.
(E) Sequence, functional and evolutionary properties significantly associated with HAP1 essential genes relative to HAP1 nonessential genes (ess/noness). * Indicates level of statistical significance (* P < 10^-3, ** P < 10^-10, Wilcoxon rank-sum).
(F) Box plot showing the conservation of HAP1 essential genes across the indicated evolutionary classes.
(G) Box plot indicating average gene expression values (log2 [TPM+1]) for the indicated gene sets broken down by fitness categories.

Fig. S3. Genome-scale genetic interaction analysis in HAP1 cells, Related to Figure 1 and STAR Methods.

(A) Diagram of genetic interaction analysis pipeline in co-isogenic cell lines. The quantitative genetic interaction (qGI) score is based on the difference between log fold change measurements for a given library gene in the query mutant (i.e., double mutant) versus WT (i.e., single mutant) cell populations.
(B) Scatterplots depicting genetic interactions for the indicated query genes. FANCG, PDCD5, VPS52 screen identifiers correspond to GIN192, GIN189, and GIN241, respectively. Negative (blue) and positive (yellow) genetic interactions that satisfied a standard genetic interaction threshold (|qGI| > 0.3, FDR < 0.1) are shown. Specific negative and positive interactions identified in each screen are indicated.
(C) Heatmap of qGI values for selected reproducible genetic interactions (columns) from biological replicate screens (n=5) for the indicated query genes (rows). Negative qGI scores are shown in blue and positive qGI scores in yellow. Genes labeled in panel B are indicated in bold face. Functions enriched among specific groups of library genes are indicated.
(D) Examples of functionally enriched gene modules derived from clustering of the entire genetic interaction dataset, as described (39). Node color represents shared general function and the poorly characterized HEATR6 gene is shown in red.

Fig. S4. Characterization of HAP1 query mutant cell lines, Related to STAR Methods.

(A) Functional distribution of 222 query gene mutant cell lines screened on a genome-wide scale for genetic interactions in HAP1 cells.
(B) Box plots showing the distribution of single mutant fitness effects for HAP1 genes grouped into the indicated fitness categories. Query genes are indicated in bold and grey box plot. The number of genes in each category is indicated in brackets.
(C) Box plot indicating average gene expression values (log2 [TPM+1]) for the indicated gene sets broken down by fitness categories. Query genes are indicated in bold and grey box plot.

Fig. S5. Data quality and reproducibility, Related to STAR Methods.

(A) Heatmap illustrating precision and recall estimates derived using an MCMC approach based on a set of screens with at least 4 replicates each. Robust estimated (i) precision and (ii) recall are plotted at 90 effect size (qGI score) and FDR cutoffs, as described. Standard (pink) and strict (purple) qGI score and FDR thresholds used to filter the raw genetic interaction dataset for large-scale data analysis and detailed mechanistic follow-up, respectively, are indicated.
(B) In total, independent replicate screens were performed for 50 query genes, each using a different batch of a lentiviral pooled TKOv3 CRISPR library prepared over the course of the project which spanned several years, which translated to qGI scores for ~766,000 independent replicate gene pairs. Biological reproducibility analysis from screens performed for the indicated query genes (range n = 2-5). Bar plots showing the average PCC based on all independent replicates for (i) LFCs/double mutant fitness, (ii) unfiltered qGI scores, and (iii) Within-Between Correlation or WBC scores. Dark grey bars indicated query genes that are highlighted in Panel C.
(C) (i) Scatter plot of qGI scores for all replicated gene pairs tested in this study. PCCs were computed after applying the indicated confidence thresholds. (ii) Scatter plots of qGI scores derived from replicated screens with the indicated query gene mutants. Correlation and WBC statistics correspond to the specific pair of replicate screens shown in the plot.
(D) Biological reproducibility analysis from screens performed using combinations of different cell line clones and growth medium for the indicated query genes.
(E) Scatter plot of qGI scores for the same gene pairs measured in different media conditions. PCCs were computed after applying the indicated confidence thresholds.
(F) Scatter plot of qGI scores between reciprocally tested gene pairs. PCCs were computed after applying the indicated confidence thresholds.
(G) Scatter plot illustrating the correlation between HAP1 single mutant fitness and genetic interaction score (qGI) for genes with self-genetic interactions. The magnitude of positive self-interactions scores is positively correlated with single mutant fitness of the query gene indicating that the vast majority of query cell lines carry complete LOF mutations in the intended query gene.
(H) Box plots showing results of re-screening 5 query genes from our genome-wide dataset using an independent CRISPR-KO library that targeted ~1,200 genes with ~30 gRNAs/gene (i.e. ~37,000 gRNAs in total)(File S6). These gRNAs were not present in the TKOv3 library but targeted genes that showed significant genetic interactions with at least one of the 5 selected query genes in the TKOv3 library. (i) Box plot shows the distribution of all unfiltered qGI genetic interaction scores derived from screens using an independent gRNA library, for groups of gene pairs that showed a negative (blue), positive (yellow) or no interaction (grey) in screens using the TKOv3 gRNA library. (ii) Box plot showing the distribution of qGI genetic interaction scores derived from screens using an independent gRNA library and involving specific query genes that showed a negative (blue), positive (yellow) or no interaction (grey) in screens using the TKOv3 gRNA library (right). This analysis recapitulated both negative and positive interactions, indicating that the genetic interactions we identified were not driven by gRNA-specific phenotypes.

Fig. S6. PTAR1 genetic interactions identified in a CRISPR-KO versus a gene trap approach, Related to STAR Methods.

(A) (i) Scatterplot of mean consensus qGI scores derived from 4 independent PTAR1 query screens with TKOv3 (GIN003, GIN043, GIN044, GIN109). Negative (blue) and positive (yellow) genetic interactions satisfying a standard confidence threshold (|qGI| > 0.3 and FDR < 0.1) are shown. Selected negative (blue) and positive (yellow) interactions are indicated. PTAR1 negative interactions identified by the gene-trap method are shown in dark blue. (ii) Venn diagram illustrating the overlap between PTAR1 negative interactions identified in this study and the gene trap analysis (Blomen et al.).
(B) (i) Functional enrichment (hypergeometric test, Benjamini-Hochberg-corrected FDR < 0.2) of genes that showed a negative interaction with PTAR1 in this study or (ii) genes that showed a negative interaction with PTAR1 after excluding genes that also showed a negative interaction with PTAR1 in the Blomen et al. gene trap study. (iii) Bar chart indicating fold-enrichment for each of the indicated gene sets, specifically for vesicle organization.

Fig. S7. Functional evaluation of genetic interaction profiles, Related to Figure 1.

(A) Schematic illustrating genetic interaction profile similarity. The set of negative (blue) and positive (yellow) interactions for a given mutant is referred to as a genetic interaction profile. Two-dimensional hierarchical clustering groups genes together based on genetic interaction profile similarity enabling identification of highly correlated groups of genes that correspond to functionally related gene modules. Genetic interactions can also be visualized as a correlation-based network connecting genes with similar genetic interaction profiles. Using a force-directed network layout, genes with highly similar genetic interaction profiles (purple lines) are placed close to each other in the network while genes with less similar interaction profiles (grey lines) are placed further apart from one another.
(B) Genes with varying degrees of genetic interaction profile similarity were evaluated for overlap with either GO biological process co-annotations or CORUM Protein complex co-annotation using precision-recall analysis. Gene pairs were sorted based on Pearson correlation coefficients, reflecting similarity in their genetic interaction profiles. Grey dashed lines show the background rate of co-annotation for the relevant set of gene pairs. Precision-recall analysis was completed separately for genetic interaction profiles derived from all library genes in the dataset as well as genetic interaction profiles excluding genes associated with highly variable single mutant fitness and/or mitochondrial genes, as indicated.
(C) (i) Matrix of 60 unique query gene mutant cell lines clustered based on pair-wise similarity of their differential gene expression profiles (red) or (ii) of their genetic interactions (purple). Two groups of genes that cluster together in both matrices and whose genetic interaction profiles were used to correct the complete genetic interaction matrix in order to construct a genetic interaction profile similarity network are indicated as Centroid 1 and Centroid 2 and described in detail in the methods section. The query gene mutant cell lines that comprise Centroid 1 and Centroid 2 are listed.

Fig. S8. Two-dimensional hierarchical clustering of the HAP1 genetic interaction dataset, Related to Figure 1.

Negative (blue) and positive (yellow) genetic interactions are shown. Rows in the matrix correspond to 17,297 genes in the TKOv3 library. Columns in the matrix represent 298 genome-scale genetic interaction screens corresponding to 222 unique query mutant cell lines. Sections (white outlines) are expanded to allow visualization of specific query-library gene-gene interactions.

Fig. S9. Annotating gene function using the HAP1 genetic interaction network, Related to Figure 1.

(A) (i) Poorly characterized genes (i.e. GO bioprocess terms + PubMed citations < 15) that localize in a specific bioprocess–enriched cluster on the genetic interaction network shown in Fig. 1A-B. Nodes are colored according to bioprocesses shown in Fig. 1B. (ii) Selected examples of re-clustered gene modules from the complete HAP1 genetic interaction matrix, where genes with the same node color have a shared function and poorly characterized genes are indicated with grey nodes and red labels. A poorly characterized library gene, HEATR6, shared interactions in common with members of the CCT chaperonin and Prefoldin complexes, suggesting that this gene may have a role in actin or tubulin folding.
(B) (i) Scatterplot of qGI scores for a HAP1 C1orf112/FIRRMM mutant query screen. The C1orf112/FIRRMM profile suggested a role for this gene in DNA damage and repair, a prediction supported by recent studies. Negative (blue) and positive (yellow) genetic interactions that satisfied a standard confidence threshold (|qGI| > 0.3 and FDR < 0.1) are shown, and specific negative and positive interactions are labeled, including the strongest negative interaction that confirmed a previously identified PICH/ERCC6L-C1orf112/FIRRMM interaction (bold font). (ii) Regions of the HAP1 genetic interaction network that are significantly enriched (hypergeometric test, Benjamini-Hochberg-corrected, FDR < 0.001) for genes exhibiting negative (blue) or positive (yellow) genetic interactions with a C1orf112/FIRRMM query mutant cell line. Enrichment was calculated using SAFE as described in the methods.
(C) The HAP1 profile similarity network provides insights into the mode-of-action of bioactive molecules. (i) Scatterplot of chemical-genetic interactions identified in the presence of 15uM NGI-1, a small molecule inhibitor of the oligosaccharyltransferase (OST) complex. For example, genes that showed sensitivity (i.e. negative chemical-genetic interactions, blue) or resistance (i.e. positive chemical-genetic interactions, yellow) to NGI-1, were enriched for roles in protein glycosylation and vesicle trafficking and localized to the corresponding functional domain region of the genetic profile similarity network. Genes that exhibited negative and positive chemical-genetic interactions with NGI-1 are indicated. (ii) Regions of the HAP1 profile similarity network that are significantly enriched for genes exhibiting negative (blue) or positive (yellow) chemical-genetic interactions with NGI-1 (right). Enrichments for NGI-1 negative or positive chemical-genetic interactions within functional domains in the HAP1 genetic profile similarity network were determined using a hypergeometric test for over-enrichment conducted for each domain, followed by a Benjamini–Hochberg correction. Fold enrichment and significance of negative and positive chemical-genetic interaction enrichment is indicated.
(D) The HAP1 genetic profile network also highlighted functions shared among different subsets of genes associated with the same disease trait or phenotype. Regions of the HAP1 genetic interaction network that are significantly enriched for (i) genes associated with a Mendelian-inherited disease (OMIM disease) or (ii) phenotype-associated GWAS. For each disease or trait gene set, a hypergeometric test for enrichment was conducted against the 17 bioprocess domains defined in the HAP1 profile similarity network, followed by a Benjamini–Hochberg correction.

Fig. S10. Genetic interaction density analysis, Related to Figure 2.

(A) Bar charts showing genetic interaction density (observed interactions/total gene pairs screened) for all library genes by category (all genes, nonessential (noness), nonessential with fitness phenotypes (noness fitness), essential) at strict genetic interaction score and significance threshold (FDR < 0.01, |qGI| > 0.6) and for library genes excluding genes with roles in mitochondrial function at the standard score and significance threshold (FDR < 0.1, |qGI| > 0.3). Negative (blue), positive (yellow) and total (grey) interaction densities, with the number of genes in each category indicated.
(B) Line plots showing average density of negative (blue) and positive (yellow) genetic interactions for (i) all library genes and (ii) excluding genes with roles in mitochondrial function relative to library gene single mutant fitness. The average negative and positive interaction density for essential genes (ess.) is also shown on the left.
(C-E) Dot plots showing functional enrichment of genetic interaction hubs. Reactome Pathway terms statistically enriched (hypergeometric test, Benjamini-Hochberg-corrected FDR < 0.2) among highly connected genes representing total (grey), negative (blue) and positive (yellow) genetic interaction hub genes were summarized according to the functional descriptions shown.
(F) A subset of essential genes participated in genetic interactions in HAP1 cells. Sequence, functional and evolutionary properties that are significantly associated with essential library genes that exhibited high genetic interaction density relative to essential genes with few genetic interactions. The analysis was performed using (i) all genes or (ii) excluding mitochondrial genes. High interaction density was defined as the top 20% of HAP1 essential genes with the highest total genetic interaction density whereas low interaction density consisted of the 50% of HAP1 essential genes with the lowest genetic interaction density. For each feature, the mean of the high-density group and the low-density group was computed, and the log2 ratio of these means is plotted. Open bars indicate relationships for which the 95% confidence interval on the correlation coefficient includes 0.

Fig. S11. Correlation analysis of genetic interaction density, Related to Figure 2.

Negative (blue) and positive (yellow) genetic interaction density was calculated for (i) all genes targeted by the TKOv3 CRISPR library, (ii) genes expressed in HAP1 cells and (iii) genes expressed in HAP1 cells without mitochondrial genes. The standard confidence threshold (|qGI| ≥ 0.3, FDR < 0.1) was applied and interaction density was computed as the percentage of observed interactions. Pearson’s correlation coefficient was used to measure associations between genetic interaction density and features that are continuous or count based. Significant (P < 0.05, using the cor.test function in R and Pearson’s correlation) non-zero correlations relationships with negative (blue) and positive (yellow) interaction density. Open bars indicate relationships for which the 95% confidence interval on the correlation coefficient includes 0. Given that analysis of different features required different statistical tests, and some features are not expected to be independent of each other, multiple hypotheses correction procedures were not applied. HI=haploinsufficiency; LOF=loss of function.

Fig. S12. Genetic interaction density associated with genome topology and duplicated genes, Related to Figure 2.

(A) (i) Genome topology map generated based on cis- and trans-chromosomal contacts from near diploid H-C datasets. Nodes are colored to represent enrichment for essential genes (magenta), nonessential genes with fitness defects (cyan), and those labeled ‘overlap’, which contain a mixture of essential and nonessential genes with a fitness defect (grey). Node size represents the number of fitness genes overlapping with trans-contacts. Node sizes are indicated by the 2 different scales. (ii) Negative (left) and positive (right) genetic interactions mapped onto the network described in (i). Nodes represent the number of genetic interactions in a particular 1 Mb bin. (iii) Violin plots indicating the number of genetic interactions in the anchor loci compared with the rest of the genome (Mann–Whitney U test, P < 4.5e-12).
(B) Bar plots depicting the average negative (blue) and positive (yellow) genetic interaction densities for individual paralog library genes with all tested query genes relative to gene family size for paralogs sharing (i) >20% or (ii) >50% sequence identity, as well as (iii) ohnologs. The numbers of interactions involving a paralog gene tested in each family size bin are indicated (* indicates level of statistical significance, * P < 0.05, ** P < 0.01, *** P < 0.001, Wilcoxon rank-sum).
(C) Bar plots indicating the average negative (blue) and positive (yellow) genetic interaction densities for individual paralog query genes versus non-duplicated query genes surveyed in this study. Query gene paralogs were defined as genes that share (i) at least 20% sequence identity, as well as (ii) ohnologs.
(D) Positive genetic interaction density among pairs of duplicated genes with increasing sequence identity (i.e. paralogs).
(E) Bar plots depicting paralog pair genetic interaction density (|qGI| >0.3, FDR < 0.1) relative to family size for paralogs sharing (i) >20% or (ii) >50% sequence identity, as well as (iii) ohnologs. Interaction density was expressed as a percentage of all tested paralog gene pairs. The numbers of paralog pairs tested in each family size bin are indicated in brackets.
(F) Histograms of negative interaction degree ratio as evidence for asymmetric functional divergence for paralog gene pairs sharing (i) >20% or (ii) >50% sequence identity, and (iii) ohnologs. The ratio is defined as the number of unique negative interactions (degree) identified for each gene of a duplicate pair with the higher degree in the numerator. Shown for comparison is another degree ratio histogram (symmetric null model) in which interactions for every duplicate pair are redistributed to either member with equal probability (grey).

Fig. S13. Relating genetic and physical interactions, Related to Figure 2.

(A) (left) Bar charts show significant fold-enrichment (P < 0.05 hypergeometric test) for all gene pairs (excluding mitochondrial genes) encoding physically interacting proteins (PPI), co-complex proteins, co-pathway proteins, or co-expressed gene pairs among negative (blue) and positive (yellow) genetic interactions defined at the indicated strict confidence threshold. (right) Bar charts show significant fold-enrichment (P < 0.05 hypergeometric test) for all gene pairs encoding physically interacting proteins (PPI), co-complex proteins, co-pathway proteins, or co-expressed gene pairs among negative (blue) and positive (yellow) genetic interactions defined at the indicated standard confidence threshold. Enrichment was measured for all gene pairs, essential gene pairs, nonessential genes with fitness phenotypes, and nonessential genes lacking a fitness phenotype. Grey bars indicate non-significant enrichment. The numbers of gene pairs tested for enrichment are indicated in brackets.
(B) (left panels) Bar graphs summarize complexes enriched for genetic interactions among members of the same complex (Within protein complex) or enrichment of genetic interactions between pairs of protein complex (Between protein complex) as the percentage of protein complexes (CORUM database) with enriched genetic interactions, categorized as all interactions (grey), negative (blue), or positive (yellow) within/between protein complexes. (right panels) Histograms illustrate purity scores indicating the proportion of negative and positive interactions within a protein complex or between a pair of protein complexes. Purity scores range from -1 (purely negative interactions) to 1 (purely positive interactions) among complex members. The dotted grey line indicates the random expectation based on a binomial distribution. Analyses include all CORUM annotated complexes or a subset of complex annotations where redundant complexes with overlapping genes were removed. In both cases, analyses were based on complexes and complex pairs with >5 tested gene pairs. Analyses were repeated after excluding genes with mitochondrial-related functions.

Fig. S14. Functional distribution of genetic interactions, Related to Figure 3.

(A) Precision-recall plots for negative (blue) and positive (yellow) genetic interactions (|qGI| >0.3, FDR < 0.1) based on co-annotation to GO biological process terms, including and excluding genes with mitochondrial-related functions. Dashed lines indicate background co-annotation rates.
(B) Genetic interaction profile–derived hierarchy schematic. Genes with highly correlated genetic interaction profiles form small, densely connected clusters representing specific pathways or protein complexes. Intermediate correlation combines sibling clusters into larger biological process–enriched clusters, which further combine into larger clusters corresponding to cell compartments. Grey scale bar indicates sibling cluster enrichment for functional terms. Analysis includes ~1600 genes with high confidence profiles, excluding genes with mitochondrial-related functions.
(C) (i) Line graph depicts density of negative (blue) and positive (yellow) genetic interactions (|qGI| > 0.3, FDR < 0.1) within a specific level of the genetic network hierarchy. Horizontal dashed lines represent background interaction density. (ii) Stacked bar chart shows functional distribution of all negative (blue) and all positive (yellow) interactions (|qGI| > 0.3, FDR < 0.1) among genes in the genetic network hierarchy, depicting percentages within clusters corresponding to cellular compartments, bioprocesses, or complexes/pathways. The combined fraction of functionally related interactions is indicated (*). (iii-iv) Bar graphs show the fraction of negative (iii, blue) or positive (iv, yellow) interactions connecting genes within the same cluster at varying functional relatedness levels.
(D) Scatterplot of mean consensus qGI scores derived from 4 independent PELO query screens (GIN289, GIN291, GIN292, GIN295) or a SKIC2 query screen (GIN415, Data File S24) with TKOv3. Negative (blue) and positive (yellow) genetic interactions satisfying a standard confidence threshold (|qGI| > 0.3 and FDR < 0.1) are shown. Selected negative (blue) and positive (yellow) interactions are indicated.
(E) MTT cell proliferation assay of Hc3716 liver cells with or without PELO following perturbation of SKIC2. Results using two independent PELO KO mutant clones are shown (*** <0.01, * <0.05 using one-way ANOVA followed by Tukey’s posthoc test).
(F) (i) Dot plots depicting yeast network density of negative (blue) and positive (yellow) nonessential gene interactions (|SGA score| >0.08, P < 0.05) within and across biological processes for 222 randomly sampled yeast query mutants. Node size reflects the fraction of interacting gene pairs observed for a given pair of biological processes; node color indicates significance above random expectation. (ii) Dot plot shows a similar analysis based on the complete set of ~4200 yeast nonessential gene query mutant strains. Nodes on the diagonal represent interactions within the same biological process; off-diagonal nodes represent interactions between processes.

Fig. S15. Genetic suppression interactions, Related to Figure 4.

(A-C) Bar graphs illustrating quantitative analysis of specific genetic suppression interactions. Query gene single mutant fitness (grey bars), library gene single mutant fitness (gene-specific colored bars), double mutant fitness (black bars) and positive genetic interaction score (qGI, yellow bars) are shown.
(D) Bar graph showing the fold enrichment of GO biological process co-annotation among gene pairs that showed positive interactions and/or suppression interactions defined at two different suppression score thresholds. Analysis with and without mitochondrial genes is shown.
(E) Box plot showing the average fraction of DepMap cell lines that depend on the indicated groups of HAP1 essential genes for viability. Numbers of essential genes tested in each group are indicated. Analysis is based on all essential genes, including essential genes with mitochondrial-related functions. Text colors relate to node color for suppression interactions shown in Fig. 4.

Fig. S16. HAP1 positive genetic interactions, Related to Figure 5.

(A) Scatter plot of human (qGI > 0.3, FDR <0.1) and yeast (SGA score > 0.08, P <0.05) positive genetic interaction densities that occur within and between corresponding bioprocesses.
(B) (i) Box plots illustrating the distribution of the ratio of positive to negative genetic interactions for all non-essential and all essential library genes with (all genes) or without (no mitochondrial genes) mitochondrial genes. (ii) Box plots illustrating the distribution of the ratio of positive to negative genetic interactions for tumor suppressor genes and all other genes with (all genes) or without (no mitochondrial genes) mitochondrial genes. See Data S19 for gene lists.

Fig. S17. Query-driven mTOR signaling effects, Related to Figure 5.

(A) (i) Heatmap depicting hierarchically clustered genetic interaction matrix comprising 85 different query gene mutant cell lines that show coherent negative (blue) and positive (yellow) genetic interactions with library genes involved in mTORC1 and mTORC2 signaling. Two inverse patterns of genetic interactions were identified: mTOR Cluster I comprised 17 query mutant cell lines with strong negative interactions with mTORC1 genes and strong positive interactions with mTORC2 genes. (ii) Schematic of mTORC1 and mTORC2 signaling pathways. Library genes are colored based on their genetic interactions with query genes belonging to mTOR Cluster I.
(B) Immunoblots and related quantitation showing phospho-AKT1S473 and phospho-AKTS240/244 effects under baseline growth conditions in HAP1 queries representative of mTOR Cluster I or II signatures. Three blots per query represent independent replicates. Blue bars indicate mTOR Cluster I and green bars indicate mTOR Cluster II. * P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001.
(C-D) Immunoblots depicting phospho-AKT1S473 and phospho-AKTS6235/236 effects following amino acid starvation and starvation>stimulation (D) and rapamycin treatment (E) in the indicated HAP1 queries representative of mTOR Cluster II signatures.
Query mutant cell lines in mTOR Cluster I displayed normal mTORC1 activity but severely compromised mTORC2 signaling. Positive interactions in Cluster I may indicate that further perturbation of mTORC2 pathway genes does not exacerbate fitness defects. A subset (~19%, 26/139, P < 4.1x10^-10, hypergeometric test) are classified as genetic suppression, identifying genes whose LOF may improve fitness of cells with reduced mTORC2 activity. Overall, mTOR Cluster I query mutants are more dependent on mTORC1 than mTORC2 signaling.

Fig. S18. Genetic interaction conservation, Related to Figure 5.

(A) (i) Bar graph showing enrichment for negative (blue) and positive (yellow) genetic interactions in yeast among conserved gene pairs scored as negative or positive in HAP1 cells. Negative interactions in HAP1 were significantly enriched in yeast; positive interactions were not. (ii) Bar graph showing enrichment for negative (blue) and positive (yellow) interactions in HAP1 among conserved gene pairs with negative or positive interactions in yeast. Negative interactions in yeast were significantly enriched in HAP1; positive interactions were not.
(B) Density plots showing fold enrichment in conservation of negative (top) and positive (bottom) genetic interactions in S. cerevisiae for 100 randomized HAP1 networks. Observed fold enrichment of real HAP1 network is indicated by a star. Bar graphs show the number of negative and positive interactions identified in both yeast and HAP1 per human gene. Query genes contributing many conserved interactions are indicated.
(C) Bar graph illustrating biological processes enriched for conserved negative genetic interactions among human-yeast orthologous gene pairs identified in HAP1 screens. Blue bars indicate significant enrichment.

Fig. S19. Conservation of PTAR1/ECM9 genetic interactions, Related to Figure 5.

(A) Bar graph showing that human gene pairs with conserved yeast orthologs (1:1 and 1 yeast: N human) that share highly similar genetic interaction profiles in HAP1 are enriched for gene pairs with high profile similarity in the yeast network. High similarity gene pairs were defined as top 5% (light purple) or top 1% (dark purple) by Pearson’s correlation coefficient in HAP1 and yeast networks.
(B) (i) Box plot of genetic interaction profile similarity (PCC) of 4847 yeast query genes with the HAP1 PTAR1 genetic interaction profile. Similarity with yeast YKT6, BET2, and ECM9 profiles is indicated. (ii) Ranked graph of 4847 yeast query genes by similarity to PTAR1 profile; top 5% (~240) were enriched for the indicated GO bioprocess term (hypergeometric test, BH-corrected).
(C) Yeast-two hybrid analysis shows physical interaction between α and β subunits of the indicated prenyltransferases; yeast Ecm9 interacts specifically with Bet2.
(D) Dual expression of human PTAR1 and RABGGTB complements lethality of ecm9Δ. Tetrad analysis shows spore germination on glucose (repressed) or galactose (induced) media; black circles indicate ecm9Δ spores, blue circles indicate rescue by human PTAR1-RABGGTB.
(E) (i) GO biological process terms enriched among human genes negatively interacting with PTAR1; (ii) yeast genes negatively interacting with ECM9; (iii) PTAR1 and ECM9 negative interactions are both enriched for Vesicle Organization.
(F) (i) Bar graph showing enrichment for negative (blue) and positive (yellow) interactions with yeast ECM9 among conserved gene pairs with negative/positive interactions with human PTAR1. Negative PTAR1 interactions are enriched for conserved orthologs showing negative interactions with yeast ECM9; positive interactions are not. (ii) Enrichment for negative/positive interactions with human PTAR1 among conserved gene pairs with negative/positive interactions with yeast ECM9; only negative interactions are significantly enriched.
(G) Serial dilution growth assays of yeast ecm9Δ abh1Δ double mutant carrying galactose-inducible expression plasmids. Growth on glucose (repressive) or galactose (inducible) shows which genes rescue ABH1 loss and restore viability of ecm9Δ.

Fig. S20. A relationship between expression dependency in cancer cell lines and HAP1 genetic interactions, Related to Figure 6.

(A) (i) Overlap of gene pairs associated with all possible ED and qGI score combinations. (ii) Enrichment for indicated functional standards among gene pairs showing specific combinations of significant ED and qGI scores. Shaded regions indicate that gene pairs with a negative ED and positive qGI (yellow), or a positive ED and a negative qGI (blue), score combinations that share the most significant gene pair overlap.
(B) Scatter plots illustrating the relationship between TAFFAZIN single mutant fitness and ABHD18 expression, UAP1 single mutant fitness and UAP1L1 expression, and PELO single mutant fitness and FOCAD expression, across a panel of DepMap cancer cell lines (22Q4). Regression lines (black dashed lines) indicate either a negative ED score & a positive genetic interaction score (qGI), or a positive ED score & a negative qGI score. Significant negative and positive genetic interactions (|qGI| > 0.3, FDR < 0.1) for TAFFAZIN, UAP1L1, and PELO query genes are indicated, with qGI scores noted.
(C) TCGA Pan-cancer analysis of co-occurring mutations with TP53. Genes with positive interactions with TP53 and significant ED scores were evaluated for enrichment for co-occurring mutations across all tumor types in TCGA. Odds ratios reflect enrichment; the left bar plot shows predicted damaging mutations, and the right bar plot shows damaging mutations plus deletions. Shading indicates statistical significance (FDR) by Fisher exact test with Benjamini-Hochberg correction.

Fig. S21. An integrated functional network based on genetic interaction and co-essentiality profiles, Related to Figure 7.

(A) Box plots showing the relationship between features and genes in significant clusters or modules derived from the genetic interaction profile similarity network that either show highly correlated DepMap co-essentiality profiles (blue bars) or do not show highly correlated co-essentiality profiles (purple bars).
(B) Enrichment of GO biological process co-annotation and PPIs among genes in the same genetic network-derived clusters that are also supported by modules derived from the DepMap co-essentiality network (blue bars) or among genes clustering based on genetic interaction profiles alone (purple bars).
(C) Scatter plot of Z-scores associated with gene modules from the complete DepMap co-essentiality network. Modules with significant HAP1 genetic interaction profile similarity are grey; others are blue. Grey dotted line indicates threshold for significant similarity.
(D) Box plots showing relationships of indicated features for DepMap co-essentiality modules that share similar genetic interaction profiles (grey) or lack strong correlation (blue).
(E) Fraction of DepMap co-essentiality gene clusters whose members are enriched for GO-BP terms or PPIs. Fractions uniquely identified in DepMap co-essentiality network (blue) or clusters sharing highly similar genetic interaction profiles (grey) are shown.
(F) Precision-recall plots for genes with similar DepMap co-essentiality profiles (blue), genetic interaction profiles (light purple), or integrated profiles (dark purple). True positives correspond to gene pairs co-annotated to GO-BP terms (top) or CORUM complexes (bottom). Grey dashed line indicates background. Inclusion of mitochondrial genes dominates DepMap co-essentiality precision; exclusion decreases performance.
(G) Comparison of individual GO-BPs or CORUM complexes captured by DepMap (blue nodes), genetic interaction profile network (light purple), or integrated network (dark purple). Axes show AUPRC values; diagonal = equivalent performance. Nodes above/below diagonal indicate stronger clustering in the corresponding network.

Fig. S22. Extreme negative synthetic lethal and positive suppression interactions, Related to STAR Methods.

Summary of total genetic interactions identified in this study and the subset of extreme synthetic lethal or genetic suppression interactions. Negative interactions are blue, positive interactions are yellow; extreme negative synthetic lethal interactions are dark blue, and positive suppression interactions are dark yellow. The right pie chart shows overlap of extreme interactions with Mendelian disease genes (OMIM) or characterized cancer driver genes (tumor suppressors).
Extreme synthetic lethal interactions correspond to HAP1-expressed, nonessential gene pairs with a negative genetic interaction (qGI < -0.6, FDR < 0.01), where the library gene single mutant fitness (LFC) > -0.5 and the double mutant fitness < -1.0. Suppressor interactions are defined as gene pairs with a suppressor score > 0.5.