Towards self-learning agents in era of high-throughput omics - PowerPoint PPT Presentation

Towards self-learning agents in era of high-throughput omics Presenter: Ameen Eetemadi Principal Investigator: Prof. Ilias Tagkopoulos 1

I use Blue Waters to: 1. Design artificial neural 2. Determine optimal networks for gene strategies to identify expression prediction next set of experiments • Thermodynamic simulations • Synthetic data generation • Deep Learning • RNA-Seq data processing • Extensive evaluations • Gaussian Processes • Extensive evaluations 2

We have entered a new era … intensity High-throughput High-performance m/z OMICS Computing Knowledge Discovery Artificial Robotic Intelligence Equipment 3

Goal: Efficient Knowledge Discovery Maximize With Minimum E. coli Bacteria knowledge about Cost figure from: https://commons.wikimedia.org Food Basic Applications Safety Science Medicine 4

The Cycle of Knowledge Discovery E. coli Experiment Learn Intelligent Lab Data Agent 5

rn dynamic program of a cell Le Learn rn dynamics of gene expression in a cell Le Learn Genetic Neural Key Features: Captures Regulatory • Network Master Relationships Regulator Models Transcription • Factor Dynamics Published at: Bioinformatics Journal, 2018 6

Learn dynamics of gene expression in E. coli Chemotaxis Transcription Regulatory Network 7

Genetic Neural Network (GNN) is 40% more accurate (for chemotaxis genes) Gene Expression (GE) Prediction Error ● ● 0.15 ● ● ● ● ● ● Mean Absolute Error ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.10 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Synthetic Data Used ● ● ● ● ● ● ● ● 10 40 70 100 Dataset size GNN LinGNN Lasso MLP RNN BiRNN 8

Genetic Neural Network (GNN) is 40% more accurate (for networks with 10-1000 genes) Gene Expression (GE) Prediction Error ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.9 ● ● ● ● ● ● ● ● ● ● ● ● ● ● Mean Absolute Error ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.8 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.7 ● ● ● ● ● ● ● ● ● 0.6 ● ● ● ● ● ● Real Data Used ● ● ● ● ● ● ● 10 200 400 600 800 1000 Network size GNN-rnd LinGNN-rnd GNN LinGNN Lasso MLP RNN BiRNN 9

The Cycle of Knowledge Discovery E. coli Experiment Learn Intelligent Lab Data Agent 10

Optimal Experimental Design for Gene Expression Prediction 11

Accelerated Knowledge Discovery 12

We have entered a new era … RNA-Seq Blue Waters intensity High-throughput High-performance m/z OMICS Computing Microarrays Knowledge Discovery Genetic Neural Networks Artificial Robotic Intelligence Equipment Optimal 13 Experimental Design

Blue Water Experience Our Workload: BW Customer Support: • Extremely parallel • Fast response • Independent small jobs • High quality Thank You! Advantages: • Extremely reliable • High availability • Comprehensive documentation 14

Acknowledgements • Ilias Tagkopoulos, PhD (Principal Investigator) • Xiaokang Wang, PhD Candidate • Navneet Rai, PhD Funding: • Beatriz Merchel Piovesan Pereira, PhD Candidate 15