Understanding Biological Function in Times of High Throughput and Low Output
1 like690 views
The document discusses the challenges and solutions in protein function annotation within the context of high-throughput biological research. It highlights issues such as high error rates in databases and the effectiveness of homology transfer, while suggesting the need for improved annotation assessment methods and more accurate software. The conclusions emphasize that combined methods yield better predictions, especially for molecular functions compared to biological processes, and stress the importance of understanding biases in current databases.
Understanding Biological Function in Times of High Throughput and Low Output
- 1. Understanding Biological Function in Times of High Throughput and Low Output Iddo Friedberg Iowa State University http://iddo-friedberg.net @iddux
- 2. Big Data in my lab Gene block evolution Images and Genomes Host/Microbiome Database error and bias Critical Assessment of Protein Function Annotations
- 3. Big Data in my lab Database error and bias Critical Assessment of Protein Function Annotations
- 4. Big Data in my lab Database error and bias Critical Assessment of Protein Function Annotations Understanding methods Understanding the data
- 5. Understanding Methods: The Critical Assessment of protein Function Annotations Pedja Wyatt Sean Tal Alex
- 6. Large Data Biology has a Bad Rap? "So we now have a culture which is based on everything must be high-throughput.I like to call it low-input, high-throughput, no-output biology" – Sydney Brenner
- 7. Motivation: The Knowledge Gap ● The gap between data and Information Information Data Temperton & Giovannoni Curr. Opin. Microbiology (2012)
- 8. Errors Accumulate in Databases Schnoes A et al (2009) PLoS Computational Biology, 5 (12)
- 9. Assigning Function to Proteins Low-ish throughput High throughput Machine learning
- 10. Most Proteins are Annotated Electronically Experimental Computational Electronic Other 0 10 20 30 40 50 60 70 80 90 100 Arabidopsis mouse Cow Zebrafish Chicken Human Compiled from the GOA project, EBI, 6/2011
- 11. Problems ● Most genes are annotated electronically ● Databases have a high error rate which is growing ● Homology transfer is less effective Solutions? Assess accuracy of annotation software Write better software
- 12. Challenges in Picking Targets ● Can't use databases: circularity problem ● Experimental groups have a small “sharing timeframe” ● Function description too vague for precise GO annotation ● There are “unknown unknowns” Choose an annotated protein Prediction method uses said annotation to predict function Circular logic... … is circular
- 13. Choosing Assessment Benchmarks Function unknown Function still unknown Function still unknown Challenge opens Submission deadline Assessment time Function unknown Function still unknown Function known Benchmark? Time
- 14. BLAST Naive Molecular Function precision/ recall
- 15. BLAST Naive Biological process precision/ recall
- 16. Case Study: hPNPase Gadi Schuster (Technion)
- 18. CAFA2 vs. CAFA1 CAFA2 was held in 2014-2015 More targets (100,00 vs. 50,000) More groups (56 vs 29) Creator:MetaPost 0.993 CreationDate:2015.09.30:1653
- 19. CAFA2 vs. CAFA1 CAFA2 was held in 2014-2015 100,000 targets 147 participants Methods have improved
- 20. CAFA Conclusions & What's Next ● Homology transfer still rules. ● Combined methods work best ● Molecular Function is easier to predict than Biological Process ● Generally, the field can use improvement ● Comparison of metrics is very much needed – Why do methods perform differently under different metrics? – Is there a “best” metric? What is “best”? ● Databases are biased
- 21. Understanding Methods: The Critical Assessment of protein Function Annotations Pedja Wyatt Sean Tal Alex
- 22. protein binding protein homodimerization activity zinc ion binding transcription activator activity chromatin binding transcription repressor activity transcription factor activity two-component sensor activity specific transcriptional repressor activity DNA binding calcium ion binding identical protein binding manganese ion binding ATP binding beta-galactoside alpha- 2,3-sialyltransferase activity magnesium ion binding enzyme binding electron carrier activity structural constituent of ribosome metal ion binding Leaf terms Molecular Function David Ream(MU) Alexander Thorman (MU) Alexandra Schnoes (UCSF) Protein Binding Activity
- 23. Annotations per article Schnoes et al PloS Comp Biol (2013)
- 24. Information is in an inverse relationship to the number of proteins annotated 1 <10 <100 ≥100 Molecular Function 1 <10 <100 ≥100 Biological Process 1 <10 <100 ≥100 Cellular Component 1 <10 <100 ≥100 Informationcontent Schnoes et al (2013) Single throughput (1 protein/study) High information (12 bits) Low information (3 bits) High throughput (≥ proteins/study)
- 25. High Throughput Experiments ● Bias our knowledge ● Bias priors for function prediction programs ● Are less informative than low-throughput experiments ● Exclusively annotate genes otherwise unknown ● Fewer $$$ ● Fast results ● Consistency The GoodThe Bad
- 26. Data that Will Drive Computation ● Whole chromosome sequencing ● Epigenomics ● Integrating images: phenomics→ genomics relationships ● Fragment-based sequencing ● Proteomics ● Metabolomics ● Documents ● Network data ● Images Current Future