If a picture paints a thousand words, can I learn from an Image?

December 3rd, 2009 by Tim Moran

Machine learning continued as a growing theme at this years HCA conference.

This first HCA conference east held in Boston, September of this year, showed promise of the increasing use of machine vision tools. These tools are making their way in to the hands of the biologist for everything from subcellular classification and pattern recognition to predictive mechanism of action based on a multivariate image output. The theme continues to grow and will be a major focus at the upcoming HCA 2010 conference in January, as is evidenced by numerous talks around the subject. Mark-Anthony Bray, Ph.D., Computational Biologist, Imaging Platform, Broad Institute, will talk on quantifying  image-based phenotypes with machine learning algorithms. Peter Horvath, Ph.D., Image Processing Scientist, Light Microscopy Centre, ETH Zurich, will also discuss  machine intelligence both for classification as well as for quality control. Pattern Recognition will be applied to Image-Based Small Molecule Screening Data by John McLaughlin, Ph.D., Scientist & Manager, Biology, Rigel Pharmaceuticals, Inc. Numerous other talks by Acclerys, Novartis and Carnegie Melon, to name a few, will also have repeating themes of learning.  I can’t help but wonder if the growth in this area is due primarily to the need or if the adoption has been increased by the growing  number of informaticians working alongside the High Content Screening biologist.

For some good background on machine learning by sure to follow Dana Honeycutt’s blog postings, here’s a link to get you started. Good Models Require Good Data October 1st, 2009 by Dana Honeycutt, Ph.D.

  • Share/Save/Bookmark

A Man with 5 Rules

August 12th, 2009 by Nancy Miller Latimer, M.S.

Chatting with Christopher Lipinski at Drug Discovery & Development Week

Some ten years ago, I first “met” the Lipinski rules in a software project.  That was my last direct “hands-on” encounter with chemistry.  At Accelrys I am the senior product manager for the Biosciences and Analytics Collections for Pipeline Pilot.  Think genomics, proteomics, sequencing, and ontologies and not chemistry!  This week I was at the DDDW show in Boston – don’t think “booth babe”.

The conference was not as busy this year as it had been in the past and it was the afternoon of the last day.  A distinguished gentleman walked up to our booth wearing a name tag of “Christopher A. Lipinski,” happy to see a fellow booth dweller.   Half in jest I asked if he might be the man with 5 rules.  Turns out he was and, boy, I was in for an intellectual treat.   That Lipinski filter came to life in a new way over the next hour or so.  I was spell bound by Dr. Lipinski’s breadth of knowledge, passion for science, and his out of the box thinking.  What I didn’t anticipate were his insights into the importance of chemistry for the biomarker and translational research space.

He was saying some really awesome things so I started writing them down.    It was hard to focus on note taking because Dr. Lipinski is an excellent speaker and very animated.  Below are a few items that I am willing to share in no particular order:

  • Translational research must have good chemistry married to good biology.
  • Your company (Accelrys) combines chemistry and biology in one software application.  If biologists are using your software to look at high throughput screening (assay) data that has associated chemical structures, they could better filter out results for poor compounds.
  • When faced with people problems (like chemistry—biology conflicts) versus technical problems—the people problems are always much more difficult to solve.
  • The people side is the most important.
  • NIH is making good strides in the dialog between chemists and biologists.
  • As soon as the biologist has an assay for a small molecule they should probe/stress test the assay with compounds known historically to cause assay problems.
  • In software for the (bench) biologist – it needs to be dead easy.  Too many peer-reviewed publications have great biology but rotten chemistry.
  • Biologically active compounds are tightly clustered in chemical space.  It is always best to look for new activity in areas of chemical space where you previously found activity.
  • It takes 10 years to “mature” a medicinal chemist.  He then becomes an expert in pattern recognition even if he can’t articulate why certain structures look better than others
  • Areas of interest
    • Stem cell (non-embryonic source)  derived screening application
    • Many previously proprietary databases are now in the public domain  (See PMID:  17897036).  These provide a great starting point for the discovery of drugs for rare diseases.

Dr Lipinski’s long and prestigious career in medicinal chemistry, assay development, computational chemistry, and now in consulting, lecturing, and as an expert witness does not look anything like retirement.  That is good news for me.

Dr. Lipinski is shown here with his rapt audience.

Dr. Lipinski is shown here with his rapt audience.

Note: Lipinski’s total number of rules actually equals 4.  His rules are known as the “Rule of Five” because each of them incorporates the number 5 in some way.  For all you literalists out there, “5 Rules” should be interpreted in this way.

  • Share/Save/Bookmark