Predicting Drug Target Interaction Using Deep Belief Network
- 1. PREDICTING DRUG TARGET INTERATION USING ADAPTIVE BELIEF NETWORK Rashim Dhaubanjar (BCT/070/129) Rupika Bista (BCT/070/131) Shiva Gautam (BCT/070/192) Sunil Bist (BCT/070/192)
- 2. INTRODUCTION • Drug development - an expensive and time-consuming process with extremely low success rate • A core problem in pharmacology is the determination of interactions between drug compounds and target proteins in order to understand and study their effects. • In this project we generalize tge applicability of the method so called PDTI for drug compounds for which no interactions are known.
- 3. PROBLEM STATEMENT • Very low known drug-target interactions • No model with maximum accuracy in drug-target interaction with lowest side-effect provision • Time complexity
- 4. OBJECTIVES • Build PDTI using Adaptive Deep Belief Network framework • Apply GPU for the parallel implementation of different phases of Deep Belief Network • Evaluate the model by applying cross-validation technique with standard dataset
- 5. Deep Belief Networks (DBN) • Stack of several RBMs together.
- 6. Restricted Boltzmann Machines activation f((weight w * input x) + bias b ) = output a
- 8. DBNs can be trained by a greedy layer-wise approach: • Train first layer using your data without the labels (unsupervised). • Freeze the first layer parameters and start training the second layer using the output of the first layer as the unsupervised input to the second layer. • Use the outputs of the final layer as inputs to a supervised layer/model and train the last supervised layer(s) (leave early weights frozen). • Unfreeze all weights and fine tune the full network by training with a supervised approach, given the pre-processed weight settings.
- 10. Feed-forward nets Information flow is unidirectional Data is presented to Input layer Passed on to Hidden Layer Passed on to Output layer Information is distributed Information processing is parallel Internal representation (interpretation) of data
- 11. Backpropagation • The Backpropagation algorithm is a sensible approach for dividing the contribution of each weight.
- 12. GPU Implementation • Training a DBN is a computationally expensive task that involves training several RBMs and may require a considerable amount of time
- 13. Solution? • GPU Parallel implementation
- 14. GPU: GeForce 840M Cores / Texture CUDA: 5.0 CUDA cores: 384
- 15. Linear Vs Matrix(GPU) 0 100 200 300 400 500 600 700 800 900 1000 Iteration 1 Iteration 5 Iteration 10 Chart Title Linear Matrix(GPU)
- 16. Linear Vs Matrix(GPU) 0 100 200 300 400 500 600 700 800 900 1000 Iteration 1 Iteration 5 Iteration 10 Linear Matrix(GPU)
- 17. TASK DIVISION
- 18. LANGUAGES USED • Java – Implementation of Algorithms • JCuda – Programming for GPU • VBA – Integration with Excel
- 19. LIMITATIONS • Protein sequence feature domain profile not used in training purpose • Only drug compound 2D substructure finger print used as pre- training dataset.
- 20. Work Complete • completed the process of algorithm development followed by testing and debugging of the developed algorithm and also implemented in code.
- 21. Work Remaining • develop a proper GUI where input will be the name of drug and the output will be interaction with protein (1 if yes and 0 if no).
- 22. THANK YOU
Editor's Notes
- #7: At node 1 of the hidden layer, x is multiplied by a weight and added to a so-called bias. The result of those two operations is fed into an activation function, which produces the node’s output, or the strength of the signal passing through it, given input x.
- #8: But in this introduction to restricted Boltzmann machines, we’ll focus on how they learn to reconstruct data by themselves in an unsupervised fashion (unsupervised means without ground-truth labels in a test set), making several forward and backward passes between the visible layer and hidden layer no. 1 without involving a deeper network. In the reconstruction phase, the activations of hidden layer no. 1 become the input in a backward pass. They are multiplied by the same weights, one per internode edge, just as x was weight-adjusted on the forward pass. The sum of those products is added to a visible-layer bias at each visible node, and the output of those operations is a reconstruction; i.e. an approximation of the original input. This can be represented by the following diagram:
- #10: Pre-training a DNN. The first row represents the first layer being trained to reproduce X, the second row is where the first layer’s weights are fixed and the second layer is trained to reproduce the output of the first layer, and so on. Finally we make use of the supervised data to learn all the weights after they have been initialised to the weights learnt in the previous step
- #12: The ideas of the algorithm is: Computes the error term for the output units using the observed error. 2. From output layer, repeat propagating the error term back to the previous layer and updating the weights between the two layers until the earliest hidden layer is reached.