High Throughput Experimentation:Computational Requirements

John M. NewsamMolecular Simulations Inc.

(A Pharmacopeia subsidiary)

“Workshop on Combinatorial Methodsfor Materials Discovery”

ATP Fall National MeetingAtlanta, GA

Wednesday November 18th 1998

Potential Hindrances?

• Patent profusion– vigilance

• Unmet expectations– set reasonably

• Infrastructure cost– hindrance for academics

• Lack of standards– premature for hardware

• Inertia– resistance to change, short-term delivery focus

High-throughput Experimentation

Lead compounds for resynthesis and secondary testing

Testing requirements drive synthesis format

Library Design

QSAR#

#Quantitative Structure-Activity relationships

Pooled, parallel or discreteSynthesis

Primary Testing Performance in specific application

Physical, mechanical etc. processingProcessing

Characterization of composition, purity, phases, structure

Analytical

Infrastructure Needs

• Vertical and horizontal integration

• Adaptable

• Modular

• Geared for huge throughput

• Broadly deployable

New 1536 well HTS Format• 1536 wells, 2 l well volume

• Corning Science Products joint design

• Automated 961536 reformatterl-level fluids dispensing

• Oxidative and evaporative loss reduced

Engineering Solution

Process and Data ManagementData BaseEngines

Oracle

MaterialsAlgorithms

Display

Statistics

User Input & Workstation Interfaces

Chemistry & Materials Input

Workstation & Oracle Forms

MaterialsSpecific Tables

Analysis, Display and Data Access

Server-basedProcessing

MolecularSimulation

Luminescence data for a library of mixed metal oxides under 254nm UV irradiation

Data from E.Danielson et al., Science 279 (1998) 831

Some Specific Technology Needs

• Hits vs misses; improvement criteria• Descriptors• Experiment decision support• Abstracted feature models (AFMs)• Process optimization• Simulation for scale-up• Sensor data (unravelling response of arrays)

• Which experiments should be done ?

Making it practical: computation

– 100 R1, 100 R2, 100 R3, 100 R4 108

– 50,000 compounds/week 40 years

• How do we manage the process ?• What knowledge do the experiments yield ?

Computation Solution

‘Hard materials’

M2

M1

X

+Temp

Scaffold

‘Soft materials’

R1

R2

R3

R4

Computation Solution

Compound library design• Library Specification

– Molecular: Product or Reaction-based– Polymers, Heterogeneous catalysts ?

• Library Design– Diversity and similarity metrics– Similarity Selection– Array and mixture design

• Library Comparison• Library Focussing

– Active site model (atomic or abstracted)– QSAR Model

C2.DiversityC2.LibCompareC2.LibSelect

World Drug Index of 35,873 compounds in a space of

principal components

Abstracted Feature Models

R.C.Willson

• Abstraction of key features

• Based on activity data

• Interesting ‘active’ definition

DescriptorsTopological

Fragments

Receptor surface

Structural

Information-content

Spatial

Electronic

Thermodynamic

Conformational

Quantum mechanical

Descriptor Families

C2.Descriptor+C2.MFAC2.QSAR+C2.Synthia

Products

Plus Molecular and Quantum Methods

Descriptors - calculable molecular attributes that govern particular macroscopic properties

Computation Solution

Available, occupiable volume & framework density descriptors

(104 zeolite and zeolite-related framework types)

Correlative methods in catalyst design: Expert systems, neural networks and structure-activity relationships , in “Advances in Catalyst Design” Catalyst Advance Program (CAP) Report, The Catalyst Group, PA; in press (1998)

Structure-Activity Relationships

C2.QSAR+C2.GA

Products

Linear regression

Stepwise & multiple linear regression

Principal components analysis

Partial least squares

Genetic algorithm

Genetic function approximation

Statistical Models

Descriptors Correlative Methods Properties

E.g. K.F. Moschner and A. Cece, “Development of a General QSAR for Predicting Octanol-Water Partition Coefficients and its Application to Surfactants,” ASTM STP 1218 (1995); MSI C2 QSAR manual April 1997.

Computation Solution

0

2

4

6

8

10

12

0 2 4 6 8 10 12

Activ ity

GF

A P

redi

cti

on

Oil Field Corrosion InhibitorsOrganics

H. Gråfen et al., Werkstoff und Korrosion, Vol. 36, 407 (1985)

• Benzimidazolines function at cathodic sites• Library studied by Kuron et al. (1985)• Key descriptors

• Terminal N charge• 3-substituted N charge• Octanol-water logP• Moment of inertia

M.Doyle

Conclusion

• Computational infrastructure needs• Specific technology needs• Role of computation

– process management system– experiment decision support– data visualization and analysis– knowledge from the experimental data

• Integration