Genevera Allen, Rice University - Stanford Big Data 2015

TL;DR

Statistical data integration is a method to combine different types of biomedical data in one joint model for more powerful and comprehensive analysis.

Transcript

hi i'm very happy to be here today um today i want to speak about data integration and data integration from a statistical perspective so at this conference we've heard so much about big data from so many different domains from genetics proteomics molecular biology medical imaging neuroimaging population health mobile health there's tons of these d... Read More

Key Insights

• 😃 Big data in biomedical research, collected from different domains and measurements, requires statistical data integration to make comprehensive inferences.
• 😫 Statistical data integration allows harnessing the power of multiple data sets for better inference on clinical outcomes and biomarker relationships.
• 🌥️ Network models, such as Markov networks, offer a valuable framework to visually represent and analyze complex biological systems in large-scale data.

Questions & Answers

Q: What is the challenge in analyzing data from the Cancer Genome Atlas?

The challenge lies in integrating multiple types of molecular profiling data, such as gene expression and somatic mutations, that are part of the same complex biological system. Current statistical and machine learning approaches analyze these data sets separately, limiting joint inferences.

Q: What are the advantages of statistical data integration in biomedical analysis?

Statistical data integration harnesses the power of multiple data sets, increasing statistical power for better inference on patient outcomes. It also allows the discovery of relationships between biomarkers that may not be apparent when analyzing data separately.

Q: How do network models, specifically Markov networks, aid in data integration?

Markov networks provide an effective tool for modeling complex biological systems by visualizing big data and capturing relationships between biomarkers. They enable the integration of various data types, including continuous, count, and binary variables, facilitating comprehensive analysis.

Q: What is the significance of block directed graphical models in statistical data integration?

Block directed graphical models allow the integration and joint analysis of mixed types of variables, including continuous, binary, and count data. This approach effectively models dependencies between different types of biomarkers and enables inference on complex biological systems.

Summary & Key Takeaways

• The wealth of biomedical data from various domains presents a challenge of integrating it into a single statistical model to make sense of complex biological systems.

• Statistical data integration aims to combine different types of data, such as gene expression, somatic mutations, and methylation, to make joint inferences and discover relationships between biomarkers.

• Network models, particularly Markov networks, provide a powerful framework to visually represent and analyze big data in biological systems.