Skip to main content

Table 4 Detection accuracies using the datasets either alone or combined

From: Quantitative maps of genetic interactions in yeast - Comparative evaluation and integrative analysis

 

Positive genetic interactions

Negative genetic interactions

Dataset pair

Early sensitivity

Partial AUC

Overall AUC

Early sensitivity

Partial AUC

Overall AUC

GIM - SGA

      

GIM rank

0.205

0.445

0.794

0.488

0.624

0.872

SGA rank

0.205

0.477

0.785

0.494

0.595

0.795

Borda count

0.432

0.527

0.826

0.562

0.619

0.857

Minimum rank

0.386

0.496

0.777

0.556

0.663

0.897

Maximum rank

0.227

0.525

0.856

0.550

0.601

0.804

Rank product

0.432

0.522

0.795

0.594

0.680

0.892

E-MAP - SGA

      

E-MAP rank

0.345

0.637

0.889

0.286

0.510

0.821

SGA rank

0.148

0.338

0.772

0.208

0.366

0.734

Borda count

0.348

0.542

0.868

0.304

0.477

0.805

Minimum rank

0.337

0.511

0.832

0.252

0.477

0.825

Maximum rank

0.255

0.575

0.887

0.294

0.460

0.769

Rank product

0.347

0.539

0.854

0.304

0.501

0.826

GIM - E-MAP

      

GIM rank

0.300

0.450

0.792

0.237

0.397

0.783

E-MAP rank

0.333

0.579

0.884

0.256

0.411

0.767

Borda count

0.367

0.533

0.878

0.293

0.457

0.807

Minimum rank

0.400

0.508

0.839

0.298

0.463

0.825

Maximum rank

0.333

0.572

0.892

0.279

0.438

0.768

Rank product

0.367

0.530

0.857

0.326

0.480

0.827

Random classifier

0.010

0.050

0.500

0.010

0.050

0.500

  1. The bolded values indicate the best ranking method in terms of the three evaluation metrics: Early sensitivity, true positive rate (sensitivity) at 1% false positive rate (FPR, 99% specificity); Partial AUC, area under the curve (AUC) at 10% FPR; Overall AUC, AUC at 100% FPR. The random classifier accuracies were used as baseline for assessing whether similar accuracies could be obtained by pure chance alone. The rankings provided by the individual datasets (GIM rank, SGA rank and E-MAP rank) were used as reference values for the integrative detection using four different rank aggregation functions (minimum, maximum, product and Borda count). The detection accuracies of the E-MAP ranking alone may be highly overestimated due to a large number of such genetic interaction pairs in the BioGRID that were extracted from the other large-scale E-MAP screening studies (see Table 2).