Protein Interface Prediction using Graph Convolutional Networks
Authors: Alex Fout, Jonathon Byrd, Basir Shariat, Asa Ben-Hur
NeurIPS 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In our experiments, several graph convolution operators yielded accuracy that is better than the state-of-the-art SVM method in this task. |
| Researcher Affiliation | Academia | Alex Fout Department of Computer Science Colorado State University Fort Collins, CO 80525 fout@colostate.edu Jonathon Byrd Department of Computer Science Colorado State University Fort Collins, CO 80525 jonbyrd@colostate.edu Basir Shariat Department of Computer Science Colorado State University Fort Collins, CO 80525 basir@cs.colostate.edu Asa Ben-Hur Department of Computer Science Colorado State University Fort Collins, CO 80525 asa@cs.colostate.edu |
| Pseudocode | No | No structured pseudocode or algorithm blocks were found in the paper. |
| Open Source Code | Yes | Python code is available at https://github.com/fouticus/pipgcn |
| Open Datasets | Yes | In our experiments we used the data from Version 5 of the Docking Benchmark Dataset, which is the standard benchmark dataset for assessing docking and interface prediction methods [25]. data can be downloaded from: https://zenodo.org/record/1127774 |
| Dataset Splits | Yes | For our test set we used the 55 complexes that were added since version 4.0 of DBD, and separated the complexes in DBD 4.0 into training and validation sets. ... Dataset sizes are shown in Table 1. Table 1: Train 140 Complexes, Validation 35 Complexes, Test 55 Complexes. ...For training we downsample the negative examples for an overall ratio of 10:1 of negative to positive examples; in validation and testing all the negative examples are used. |
| Hardware Specification | Yes | Training times vary from roughly 17-102 minutes depending on convolution method and network depth, using a single NVIDIA GTX 980 or GTX TITAN X GPU. |
| Software Dependencies | Yes | We implemented our networks in Tensor Flow [1] v1.0.1 |
| Experiment Setup | Yes | The validation set was used to perform an extensive search over the space of possible feature representations and model hyperparameters, to select the edge distance feature RBF kernel standard deviation (2 to 32), negative to positive example ratio (1:1 to 20:1), number of convolutional layers (1 to 6), number of filters (8 to 2000), neighborhood size (2 to 26), pairwise residue representation (elementwise sum/product vs concatenation), number of dense layers after merging (0 to 4), optimization algorithm (stochastic gradient descent, RMSProp, ADAM, Momentum), learning rate (0.01 to 1), dropout probability (0.3 to 0.8), minibatch size (64 or 128 examples), and number of epochs (50 to 1000). For testing, all classifiers were trained for 80 epochs in minibatches of 128. Rectified Linear Units were employed on all but the classification layer. During training we performed dropout with probability 0.5 |