Eric Leung Code and Data Learnings     about     blog     projects     misc     feed

Translational Medicine Talk Summary

I recently attended the joint conference between International Conference on Intelligent Systems for Molecule Biology and European Conference on Computational Biology in Dublin, Ireland. One talk I went to was Winston Hide’s talk “Translational Medicine: The Current Landscape and Future Directions.” The following is my summary of that talk, along with some personal commentary from me.

On a quick side note, Hide mentions that Russ Altman’s annual Translational Bioinformatics Year in Review talk has influenced his talk. Here is an awesome list of links from Altman’s talk.

Brief Intro to Translational Medicine

One definition of “translational medicine” is the combination of scientific discovery and health improvement. Another succinct way to put it is a “bench-to-bedside” approach, taking research findings at the “bench” and applying it to the “bedside” where it affects patients, public health, and others fields such as policy. This is a very powerful approach and as Winston Hide puts it, “we’re dealing with people, bringing quantitative information to people.”


There were up’s and down’s to translational sciences through the 1950’s and onward. The three up’s were induced by pharmacology, affinity assays, and monoclonal antibodies.

Drugs and pharmacology are the first things that come to mind when thinking about bench-to-bedside. You start off finding a binding site of a protein, which when blocked can decrease the amount of a disease inducing agent. What now? Design a drug for it! This boom of translational sciences occurred in the 1950’s using pharmacological, biochemical, and physiological approaches.

The next wave of innovation was in the 1980’s with affinity assays. My guess is that these are referring to ligand binding assays. One possible invention was Partition Affinity Ligand Assay (PALA).

Lastly, before getting to where we are now with high throughput sequencing, there was another upswing due to monoclonal antibodies, as opposed to polyclonal antibodies. Monoclonal antibodies are amazing because they bind to single antigen type, which make them very specific in targeting. One example of using monoclonal antibodies is in cancer drugs. Here is more on the difference between monoclonal and polyclonal antibodies.

The Tsunami of Data

There was a time when there was just THE human genome project, trying to sequence THE genome. Nowadays, it is all about sequencing MANY whole genomes. To make this data tsunami even larger, there are complementary pieces of information that go with the genome analyses (such as annotation data from ENCODE and patient information) which make the analyses more difficult. The exciting, rugged sea of computational analyses also yearns for a bit of oversight in order for analyses and infrastructure to be scalable and stable. In time, we may have the problem of too much data and not enough resources to store and analyze it.

Biomedical Discovery

Lee Hood was quoted saying something along the lines of “biomedical discovery is becoming a data science.” This was back in 1993 and he also commented on how around 20% would involve data. Times have changed and that percentage is much larger. Lee Hood went off to co-found the Institute for Systems Biology in 2000 and since then has made great strides in data driven research.

IT in Hospitals

In the United States, there has been a shift from using paper health records to electronic ones, part of which was facilitated by government policy. The days we dream of where health IT can efficiently and confidently improve healthcare is still but a dream. It is much easier to start from scratch but it is difficult to just changed the workflows of a hospital full of healthcare teams.

Population Sequencing

Recently, there was a Nature article on the large-scale whole genome sequencing of the Icelandic population. First was sequencing one genome, next is sequencing many genomes, lastly is sequencing populations and communities of genomes to gain insight into human disease and health. Once this data from the Icelandic population is available to use, coupling it with individual genome analysis is bound to give great insights.

Consumer Genetic Testing

Sequencing is done on populations because governments will invest the high cost for sequencing. However, sequencing is becoming cheap enough to market to the general public with direct-to-consumer genetics testing. Here’s a comprehensive analysis of these genetic tests. Some such companies are:

Do These Genetic Tests Work?

Recently, there was a New York Times article by Kira Peikoff on direct-to-consumer genetic tests and their disagreement with each other’s test results. She got tests from three different companies and got different answers. Why? The lack of solid evidence to make these claims but also the fact that companies are only performing spot variant checking (i.e. companies choose which variants to look at), instead of looking at the whole genome. The analogy they used in the article was “Imagine if you took a book and you only looked at the first letter of every other page,” said Dr. Robert Klitzman.

Genes to Drugs Paradigm Shift

In the past, genetic analyses were performed on single genes. From the insights on these genes, drug targets were identified to make a drug from. However, in practice, this drug would fail because of underlying interactions the gene in involved in which make the drug fail. In other words, developing a drug in this way doesn’t account for the body’s ability to maintain equilibrium.

Thus, there is a paradigm shift in thinking of networks of genes, rather than the individual genes separately. This has allowed more complex understanding, in theory, of a drug’s effectiveness with a larger picture of what is really going on. You can even add genetic, information from model organisms, and expression data to broaden the picture even more.

Accelerating Medicines Partnership

It is great if you can create a drug and maybe it’ll work. However, that is just one piece of the puzzle. There are funders and other drug companies. The funders want people to work together so they can efficients invest their money. Meanwhile, the drug companies want to minimize as much risk as possible, as developing drugs and getting it past the three phases of drug design is a long process. Thus, the Accelerating Medicines Partnership was founded as a public-private partnership between NIH, FDA, ten biopharmaceutical companies, and others. They plan on tackling Alzheimer’s disease, type 2 diabetes, and lupus.

Genes Expression and Networks

Coming back to genes and networks, you can visualize and understand a much larger picture of what is going on in the body when you look at networks of networks. Similar to a gene co-expression network where you join points in a graph if they interact, you can create what is called a pathway co-expression network whereby a collection of gene interactions (i.e. pathways) interact with each other. This allows you to get at a functional model of if this pathway of genes goes up, this other pathway goes down.

Common Cloud Storage for Analysis

Data analysis and knowledge comes from the concerted efforts of everyone. With large datasets that are the subject of analyses, sharing the analysis can’t be as simple as emailing or transferring the data to a colleague.

There has been discussions about using the cloud to alleviate this collaboration problem, but also to make the analyses more efficient. According to that linked discussion by Nature costs more ($18 on the cloud vs $200 on academic data centers) and takes more time (6 weeks on the cloud vs 6+ months in data centers). Some companies have made an effort to make these analyses faster and more accessible.


Hide finishes with some things to look out for in the future: a data patient, proactive treatments, machine learning treatments, interplay between genetics and environment, and predict genetics.

The future may hold a time where patients and their treatments will be data driven whereby as the disease is progresses, it can be analyzed in real-time so that treatments can be effectively developed.

A step before that may be improving predicting disease incidence and making preventive measures for the disease to develop. Along the lines of prediction, smart and trained systems, such as supercomputer Watson, can be used to crunch massive amounts of data and make informed advice on treatments to assist physicians understand the plethora of data on their patients.

Lastly, we hope to understand the role the environment plays on your genome to create who and what you are. Predictive simulations might also be developed to guess what you are based solely on your genome.


The field of translational medicine is still in its infancy. There are hurdles in scaling analyses and making it easier available to the public but standards for interpretation and governance may help solve the problem. Partnerships between industry and government agencies also help create a sustainable model for genomic analyses to help improve human health.