Skip to main content

Table 2 Results of the SAGE models for the goal-directed benchmarks in GuacaMol

From: Development of scoring-assisted generative exploration (SAGE) and its application to dual inhibitor design for acetylcholinesterase and monoamine oxidase B

Task

Target Compound

SMILES LSTMa

SMILES GAa

M100

(GEGL)

M75/V20/B05

M50/V45/B05

M25/V70/B05

Rediscovery

Celecoxib

0.609

0.506

1.000

1.000

1.000

1.000

Troglitazone

0.465

0.333

1.000

1.000

1.000

1.000

Thiothixene

0.544

0.433

1.000

1.000

1.000

1.000

Similarity

Aripiprazole

0.740

0.609

1.000

1.000

1.000

1.000

Albuterol

0.835

0.644

1.000

1.000

1.000

1.000

Mestranol

0.822

0.466

1.000

1.000

1.000

1.000

Isomer

C11H24

0.716

0.872

1.000

1.000

1.000

1.000

C9H10N2O2PF2Cl

0.738

0.769

1.000

1.000

1.000

0.984

Multiple Property Optimization

Fexofenadine

0.794

0.752

1.000

1.000

1.000

1.000

Ranolazine

0.799

0.758

0.948

0.948

0.955

0.951

Perindopril

0.556

0.519

0.848

0.882

0.845

0.883

Amlodipine

0.690

0.622

0.906

0.913

0.924

0.924

Sitagliptin

0.494

0.470

0.912

0.926

0.929

0.925

Zaleplon

0.540

0.510

0.788

0.838

0.795

0.792

Valsartan SMARTS

0.334

0.027

0.999

1.000

0.997

0.999

Decorator Hopping

0.912

0.624

1.000

1.000

1.000

1.000

Scaffold Hopping

0.670

0.525

1.000

1.000

1.000

1.000

Total

11.258

9.440

16.401

16.507

16.443

16.458

  1. M is a mutate operator, V is a virtual synthesis operator, and B is a bridged bicyclic ring operator
  2. aWithout the ranking-based fine-tuning mechanism
  3. Bold value indicates the best results