Skip to main content

Table 3 Results of the Buchwald-Hartwig dataset under ligand-based out-of-sample conditions

From: Prediction of chemical reaction yields with large-scale multi-view pre-training

Split type

Measure

YieldBERT

YieldBERT-DA

UA-GNN

ReaMVP

Test 1

MAE

\(7.351\pm 0.099\)

\(\mathbf {7.015\pm 0.758}\)

\(8.082\pm 0.827\)

\(7.276\pm 0.124\)

 

RMSE

\(11.441\pm 0.342\)

\(11.761\pm 1.398\)

\(13.746\pm 1.175\)

\(\mathbf {10.768\pm 0.136}\)

 

R\(^2\)

\(0.824\pm 0.010\)

\(0.811\pm 0.047\)

\(0.744\pm 0.042\)

\(\mathbf {0.844\pm 0.004}\)

Test 2

MAE

\(7.266\pm 0.724\)

\(6.588\pm 0.328\)

\(6.300\pm 0.647\)

\(\mathbf {6.078\pm 0.149}\)

 

RMSE

\(11.144\pm 1.267\)

\(9.886\pm 0.741\)

\(9.476\pm 1.027\)

\(\mathbf {8.722\pm 0.179}\)

 

R\(^2\)

\(0.829\pm 0.037\)

\(0.866\pm 0.020\)

\(0.876\pm 0.026\)

\(\mathbf {0.896\pm 0.004}\)

Test 3

MAE

\(9.129\pm 0.745\)

\(11.052\pm 0.950\)

\(8.986\pm 0.314\)

\(\mathbf {8.969\pm 0.491}\)

 

RMSE

\(14.276\pm 0.820\)

\(18.041\pm 1.395\)

\(14.939\pm 0.622\)

\(\mathbf {12.791\pm 0.769}\)

 

R\(^2\)

\(0.741\pm 0.030\)

\(0.585\pm 0.067\)

\(0.717\pm 0.024\)

\(\mathbf {0.792\pm 0.025}\)

Test 4

MAE

\(13.671\pm 1.067\)

\(18.422\pm 0.620\)

\(13.190\pm 0.754\)

\(\mathbf {10.605\pm 0.656}\)

 

RMSE

\(19.679\pm 1.397\)

\(24.279\pm 0.494\)

\(18.774\pm 0.566\)

\(\mathbf {14.618\pm 0.932}\)

 

R\(^2\)

\(0.444\pm 0.077\)

\(0.157\pm 0.034\)

\(0.496\pm 0.031\)

\(\mathbf {0.693\pm 0.038}\)

Plate 1

MAE

\(10.036\pm 0.300\)

\(\mathbf {8.880\pm 0.552}\)

\(10.981\pm 0.624\)

\(9.576\pm 0.299\)

 

RMSE

\(14.832\pm 0.367\)

\(\mathbf {13.697\pm 0.432}\)

\(15.467\pm 1.045\)

\(13.808\pm 0.372\)

 

R\(^2\)

\(0.752\pm 0.012\)

\(\mathbf {0.789\pm 0.013}\)

\(0.730\pm 0.037\)

\(0.785\pm 0.011\)

Plate 2

MAE

\(16.822\pm 1.988\)

\(\mathbf {14.449\pm 0.375}\)

\(15.547\pm 1.004\)

\(14.651\pm 1.777\)

 

RMSE

\(21.711\pm 2.283\)

\(19.682\pm 0.342\)

\(21.479\pm 1.617\)

\(\mathbf {19.356\pm 2.003}\)

 

R\(^2\)

\(0.181\pm 0.171\)

\(0.334\pm 0.023\)

\(0.202\pm 0.121\)

\(\mathbf {0.349\pm 0.129}\)

Plate 3

MAE

\(9.932\pm 0.287\)

\(10.796\pm 1.016\)

\(\mathbf {8.163\pm 0.570}\)

\(8.855\pm 0.537\)

 

RMSE

\(13.714\pm 0.341\)

\(14.788\pm 1.287\)

\(\mathbf {11.901\pm 0.635}\)

\(12.139\pm 0.479\)

 

R\(^2\)

\(0.718\pm 0.014\)

\(0.669\pm 0.056\)

\(\mathbf {0.787\pm 0.023}\)

\(0.779\pm 0.017\)

Plate 2 new

MAE

\(12.629\pm 1.259\)

\(11.521\pm 0.495\)

\(12.546\pm 1.071\)

\(\mathbf {10.322\pm 0.556}\)

 

RMSE

\(17.509\pm 1.917\)

\(16.540\pm 0.271\)

\(18.568\pm 1.387\)

\(\mathbf {13.987\pm 0.583}\)

 

R\(^2\)

\(0.508\pm 0.106\)

\(0.566\pm 0.014\)

\(0.451\pm 0.083\)

\(\mathbf {0.689\pm 0.026}\)

  1. Bold entries highlight the best performance