Hot Topics in Chemoinformatics in the Pharmaceutical Industry

David J. Wild, Ph.D.

Scientific Computing Consultant, and Adjunct Professor of Pharmaceutical Engineering at the University of Michigan
david@wild-ideas.org www.WildIdeasConsulting.com

About me
B.Sc Computer Science Ph.D. Chemoinformatics (Willett Lab) Worked for 5 years in Scientific Computing leadership at Pfizer, responsible for the development of computational tools for scientists Now run a consulting firm based in Ann Arbor, Mich., and am also an Adjunct Professor at the University of Michigan.

doing some research

Wild Ideas Consulting www.WildIdeasConsulting.com

University of Michigan www-personal.engin.umich.edu/~wildd

What well cover today

Overview of early-stage drug discovery and the big industry concerns Using information and technology together to improve the chances of finding a new drug Example High Throughput Screening Some other examples of hot areas
Genomics & Proteomics Information Handling Virtual Screening Combinatorial Chemistry Design of scientific software

Characteristics of the pharmaceutical industry

Very segmented market largest company (Pfizer) only has an 11% market share High risk, long term takes 10-20 years to develop a drug, and most drugs fail to get to market Highly regulated (by FDA) High profit margins for drugs which do make it Investors traditionally expect high return on investment Four main phases: discovery, development, clinical trials and marketing

R&D spending up, new drugs down

Taken from http://www.newscientistjobs.com/biotech/ernstyoung/blues.jsp

Drug Discovery & Development

Identify disease Find a drug effective against disease protein (2-5 years) Isolate protein involved in disease (2-5 years)

Preclinical testing (1-3 years)

Formulation & Scale-up

Human clinical trials (2-10 years)

FDA approval (2-3 years)

Impact of new technology on drug discovery

The last few years have seen a number of revolutionary new technologies:
Gene chips, genomics and HGP Bioinformatics & Molecular biology More protein structures High-throughput screening & assays Virtual screening and library design Docking Combinatorial chemistry In-vitro ADME testing Other computational methods

How do we make it all work for us?

GENOMICS, PROTEOMICS & BIOPHARM.

Potentially producing many more targets and personalized targets

HIGH THROUGHPUT SCREENING

Identify disease Screening up to 100,000 compounds a day for activity against a target protein

VIRTUAL SCREENING
Isolate protein Using a computer to predict activity

COMBINATORIAL CHEMISTRY
Rapidly producing vast numbers of compounds Find drug

MOLECULAR MODELING
Computer graphics & models help improve activity

IN VITRO & IN SILICO ADME MODELS

Preclinical testing

Tissue and computer models begin to replace animal testing

There is little hard data on using the new technologies

In a sense, the drug design process is becoming a big experiment Do we continue as before, and carefully introduce new technologies as we deem appropriate, or do we radically change the way things are done? Lots of pressure for the new technologies to yield results quickly How do we measure the results?

Some questions being asked

Is our increasing spending on R&D and new technologies really going to pay off? Or was it a red herring? Is the paucity of drugs in the pipeline because were not doing things right, or are we just hitting limits on the number of major diseases with potential treatments still to be found? (all the low-hanging fruit has gone) Should we be looking in new areas (e.g. life enhancment rather than life saving or quality of life)

Whats being done

Trying to get the right Attrition (=drugs dropping out of the pipeline). Aim to increase early-stage attrition and reduce late-stage attrition Risk analysis look ideally for low-risk, high-payoff drugs Using metrics to monitor successes and failures

Analyzing risk

High risk Low payoff

High risk High payoff

Low risk Low payoff

Low risk High payoff

Using metrics to monitor improvement

Split the discovery process into discrete units, with key decisions at the end of each unit. Come up with measurable properties that can be used to gauge success Look for good and bad decisions and why they were made Stage Target exploration HTS HTS Analysis Decision Point Go with this target? Was the screen successful? Follow up these 5-10 series

Series Followup Produce 2-3 lead compounds ADME study Are compounds safe?

Summary
The pharmaceutical industry is a high-risk industry with very long development times and short product lifespans There has been a lot of investment in new technologies for early stage drug discovery, but so far these are not resulting in more drug candidates (or profits) Companies are looking at ways to address this problem including managing attrition, risk analysis and metrics.

How Chemoinformatics can help out

Producing and manage information for metrics In-silico analysis to reduce risk, e.g.
Virtual screening Library design, Docking Cost/benefit analyses

Making information available at the right time and the right place Needs to be integrated into processes

An example: High-Throughput Screening

Screening perhaps millions of compounds in a corporate collection to see if any show activity against a certain disease protein

High-Throughput Screening
Drug companies now have millions of samples of chemical compounds High-throughput screening can test 100,000 compounds a day for activity against a protein target Maybe tens of thousands of these compounds will show some activity for the protein The chemist needs to intelligently select the 2 - 3 classes of compounds that show the most promise for being drugs to follow-up

Informatics Implications
Need to be able to store chemical structure and biological data for millions of datapoints
Computational representation of 2D structure

Need to be able to organize thousands of active compounds into meaningful groups

Use cluster analysis or machine learning methods to group similar structures together and relate to activity

Need to learn as much information as possible from the data (data mining)
Apply statistical methods to the structures and related information

HTS Tools Tripos SAR Navigator

SAR Navigator is Tripos, inc., www.tripos.com

BioReason ClassPharmer
Clusters actives into groups representing series Attempts to find a scaffold using MCS algorithm Recovers inactives back into series Presents series as rows in a spreadsheet view Gives other statistics on series, such as activity distribution http://www.bioreason.com

BioReason Classpharmer

www.bioreason.com

BioReason Classpharmer

www.bioreason.com

Strategy for HTS Triage

Run HTS Decided which compounds are active and which are inactive Cluster the actives to put them into series Visualize clusters of actives (showing 2D structures) and pick series of interest Identify scaffold for each series Use similarity or substructure search on inactives to find inactives related to these series Use SAR techniques to discover differences between actives and inactives in a series

Information generated at different points in the Drug Design process

Gene chip experiments Protein structures Project selection decisions Assay protocols HTS results Series selection decisions SAR studies Combinatorial Expts. Pharmacophores ADME studies Lead cmpd decisions Toxicology studies Scaleup reactions

Clinical Trials data Doctor/patient studies Marketing, surveys, etc

Information generated at different sites

Distributed goals model

Shared goals model

Information storage breakdowns

Large amounts of information generated:
Some is not kept at all Some is kept but loses its meaning

Often data is kept, but not semantics or decisions

e.g. keep the HVX2 assay result for this compound was 3.2, but not what the assay protocol was, whether the compound was considered active, nor whether it was followed up on.

Bigger picture or derivative information is usually not stored

E.g. all the compounds with a tri-methyl group seemed to have much lower activity for this project

Information access breakdowns

Some information is only available in one physical location Some information is only available within one part of the discovery process Often information is not contextualized for use outside a particular domain When someone is clear about a piece of information they need; that piece of information exists, but they dont know how to access it.
E.g. What system to use, what Oracle table its in, or even the knowledge of whether that piece of information does exist!

Missed opportunities
Not a specific breakdown, but if the right piece of information had been available at the right time, better decisions could have been made E.g.
A group of compounds is being followed up as potential drugs, but a rival company just applied for a patent on the compounds A large amount of money is being spent developing an HTS assay for a target, but marketing research shows any drug is unlikely to be a success A group of compounds is selected from an HTS as good candidates for follow up, but 20 years ago they were followed up for a similar project and had severe solubility problems

Information use breakdowns

The meaning of data is incorrectly interpreted A single piece of information is used, whilst using a wider range of information would lead to different conclusions Lessons learned from one project are incorrectly applied to another Fuzzy information is taken as concrete information

What do we do?
No large company has really solved the problem But ongoing attempts include:
Defining information produced and needed at each stage of the discovery process Improving processes to be more consistent, especially across different sites Improving information flow between departments and sites Harmonizing terminology across disciplines and sites Defining needed management information as well as raw data Looking for quick win opportunities

This will presumably impact what is stored in databases and what software is used
Oracle Chemistry Cartridges help

Some Other Examples

Genomics & Proteomics Information Handling Virtual Screening Combinatorial Chemistry Design of scientific software

Genomics & Proteomics Information Handling

Understanding the link between diseases, genetic makeup and expression of proteins

Genomics
Genomics is fast-forwarding our understanding of how DNA, genes, proteins and protein function are related, in both normal and disease conditions Human genome project has mapped the genes in human DNA Hope is that this understanding will provide many more potential protein targets Allows potential personalization of therapies

ATACGGAT TATGCCTA

functions

Gene Chips
Gene chips allow us to look for changes in protein expression for different people with a variety of conditions, and to see if the presence of drugs changes compounds administered that expression Makes possible the design of drugs to target different phenotypes
expression profile (screen for 35,000 genes) people / conditions e.g. obese, cancer, caucasian

Chemogenomics from Vertex

Video: http://www.vrtx.com/Chemogenonone.html

Virtual Screening
Build a computational model of activity for a particular target Use model to score compounds from virtual or real libraries Use scores to decide which to make, or pass through a real screen

Computational Models of Activity

Machine Learning Methods
E.g. Neural nets, Bayesian nets, SVMs, Kahonen nets Train with compounds of known activity Predict activity of unknown compounds

Scoring methods
Profile compounds based on properties related to target

Fast Docking
Rapidly dock 3D representations of molecules into 3D representations of proteins, and score according to how well they bind

Present molecules to model

We may want to virtual screen
All of a companys in-house compounds, to see which to screen first A compound collection that could be purchased A potential combinatorial chemistry library, to see if it is worth making, and if so which to make

Model will come out with with either prediction of how well each molecule will bind, or a score for each molecule

Combinatorial Chemistry
By combining molecular building blocks, we can create very large numbers of different molecules very quickly. Usually involves a scaffold molecule, and sets of compounds which can be reacted with the scaffold to place different structures on attachment points.

Example Combinatorial Library

Scaffold R-groups
R1 = OH OCH3 NH2 Cl COOH NH
CN

Examples
OH NH

R1

O OH C

OH

R2 R3

R2 = phenyl OH NH2 Br F CN R3 = CF3 NO2 OCH3 OH phenoxy

OH

OH NH

NH

OH
CF3 O CH3 O C NH OH

For this small library, the number of possible compounds is 5 x 6 x 5 = 150

Combinatorial Chemistry Issues

Which R-groups to choose

Which libraries to make

Fill out existing compound collection? Targeted to a particular protein? As many compounds as possible?

Computational profiling of libraries can help

Virtual libraries can be assessed on computer

Design of Scientific Software

Problems with scientific software tend to occur because of deficiencies in one of three areas: Software Relevance Software Usability Software Management

Software Relevance
To be able to make software relevant requires the software designer to understand:
the science, i.e. the domain the scientific computing techniques that are used in the domain the possibilites and limitations of software development.

Even with this, its hard to match the things we can do with the things that people want or need to do Techniques like personas and contextual inquiry simply help us understand the people who use the software, their goals, and tasks they want to do

Software relevance: Bridge between computation & science

clustering sim. searching activity models scaffold detection docking logp calculation

goals: e.g. produce compounds that have high biological activity tasks: work out a chemical synthesis

tasks: doing a cluster analysis identifying activity-related fragments

tools

choose good reagents

try and document some reactions

chemoinformatics

science

Software Usability
Tend to focus on the method and the science, but not how easy it is for people to get their job done using the software Programmers tend to make software intuitive for them, but not necessarily the people it is designed for A usability lab and other techniques can make a HUGE difference to the satisfaction of users and programmers alike!

Software Management
Disparate set of tools & platforms Disparate programming styles, languages A variety of people tend to be writing software
Trained software developers Enthusiastic scientists Scientific computing specialists

Focus on the science tends to mean software management is neglected Everyone hates traditional software management rules But there are ways of making everything work better and having more fun doing it! Have a recommended basic setup that should help a lot

Foundation reading
The Inmates Are Running the Asylum by Alan Cooper Contextual Design by Hugh Beyer and Karen Holtzblatt Usability Engineering by Jakob Nielsen The Visual Display of Quantitative Information by Edward Tufte Dont Make Me Think! by Steve Krug

See also, www.WildIdeasConsulting.com

Summary
R&D in the pharmaceutical industry is undergoing a lot of technological changes, and there is pressure to make the investment pay off There is a big need to sensibly use the large amounts of chemical and biological-related information produced in the process Thoughtful use of chemoinformatics methods and software is becoming crucial to the success of drug discovery