Genzyme Process Data

Download Report

Transcript Genzyme Process Data

Heads Up Analysis Vs High Dimensionality
Problems: Using Data Visualization as an
Analytical Tool In Advanced Solutions Applied to
Manufacturing Data
Gloria Gadea-Lopez, Ph.D., Genzyme Corporation
Robert H. McCafferty, Curvaceous Software
International Forum for Process Analytical Chemistry
January 15, 2004
IFPAC 2004
Application of Parallel Cordinates as
Data Mining Tool at Genzyme
Gloria Gadea-Lopez, Ph.D.
Introduction

Genzyme Corporation is the leading
manufacturer of sterile Sodium Hyaluronate
 Unique viscoelastic properties that make it the
ingredient of choice for ophtalmic applications
and post-surgery antiadhesion products
 Produced under a proprietary Genzyme
fermentation and purification method that
yields highly purified, exceptional quality
material.
Sodium Hyaluronate (HA)


Produced by Genzyme’s
Advanced Biomaterials since
1984
In its natural form, HA typically
exists as a sodium salt (sodium
hyaluronate), which can form a
highly viscous fluid (viscoelastic)
with exceptional lubricating
qualities.

Plays an important role in a
number of physiological functions
including, cells protection and
lubrication, maintenance of the
structural integrity of tissues,
transport of molecules, and fluid
retention and regulation.
Sodium Hyaluronate - Applications

Genzyme's Seprafilm®, which is used during surgery as an adjunct
intended to reduce the incidence, extent, and severity of post-surgical
adhesion formation in the abdominopelvic cavity.
 Genzyme's HA is also being used in commercially available ophthalmic
products.
 Established and potential applications for HA include:











Ophthalmology
Soft tissue implants
Wound healing
Viscosupplementation of joints
Bone regeneration
Surface coatings
Moisturizing agents
Adhesion prevention
Cell preservation
Drug delivery
Immunomodulation
Making More and Better HA

Process with a lot of history

Existing legacy of previous facility

Lessons to be applied to new expanded facility
 Exercise with Curvaceous Software

Data from 150 batches, 70 variables
•
•
•
•
Raw material lot information
Process conditions for each unit operation
Cycle times per batch
Properties of the final product (Quality Control data for
release).
Using Parallel Coordinates - Objectives

Discover “hidden” relationship among process
variables that lead to


Higher yield
Viscoelastic properties in optimum range

Prevent process conditions in ranges that lead to
adverse results
 Optimize raw material properties

Work with suppliers to ensure consistent properties of
key raw materials.
Collecting the Data
 Information



from diverse sources
Batch records
Raw material information
Final QC (Certificate of Analysis)
 Process
history resides in MERLIN,
Genzyme’s custom process database.
Biotech Processes are not “Flat
File” Friendly
Fill/Finish
Purification
Purification
Recovery
BioReactor
LOT301
LOT401
LOT302
LOT402
LOT303
LOT403
LOT201
Final
Product
QC Data
LOT101
LOT202
Lot Pooling
Lot Splitting
In-Process Data
Raw Materials
Effects on
Final
Product
Quality?
Typical Data Analysis
 Using
ODBC, link Merlin and JMP (SAS),
Excel, Chart FX, QC Charter




Linear regression models
Process Capability
Multivariate with some key properties
Control Charts (Shewhart)
 Need
comprehensive analysis. We know
there is more……
What N-Space Really Looks Like… And
How To Make It Pay In Spades
Robert H. McCafferty, Curvaceous Software
Limited
Beyond The Third Dimension

Typical Industry Practice






Few High Return Processes Fully Understood
Complex Chain/Hierarchy Of Intricate Unit Processes
Brute Force Numerical Analysis Characterization Method Of
Choice
Human Intelligence Relegated To Back Seat
Jungle Of N-Space Impenetrable
New Process Knowledge Latent In Existing Data




Key To Extraction Engaging Human Mind… Native Curiosity
Eyes Primary Path Of Information Input To Human Brain
N-Dimensional Visualization Breakthrough Technology
3-Dimensional Status Quo Must Be Broken
Parallel Coordinates

Substantial Foundation In N-Dimensional Geometry

Map N-D Into 2-D Through Coordinate Transform

Allow Direct Data Visualization And Manipulation




Many Process Variables Simultaneously (30+)
Mathematically Robust… Zero Information Loss
No Derived Quantities (Re, Nu, PC, etc.) Required
True Visualization


Otherwise Unobservable Phenomena Easily Seen
Readily Explained
A Single 16-Dimensional Point In Parallel
Coordinates
Visual Analysis

Patterns Formed When Many Points Plotted

Human Brain Superlative Pattern Recognizer



Very Good At Seeing The Big Picture
Eyes Better Than Algorithms
No Absolute Requirement To Understand



Process Physics
Mathematics
Statistics

Process Knowledge Real Key

Anyone Can Contribute
Applying Good, Better, Best Criteria Uncovers Patterns



Sweet Spots - Concentrations Of Yellow - Appearing
High X14 (Biosynthesis End Criteria) Clearly Best For Product
Nearly All Premium Production From One X3 Level… High X5
Best
Delving Into More Variables...
Curious “Hole” In Two Process Variables
 Obvious Bad Range - Poison Zone - For Another
 Clear Relationship Between Two Others, High Values Favored

Looking At Final Process Results



Different Modes Of Process Operation Plainly Visible
Clear Sweet Spot Relationships For Key Time Variables
Fair Correlation Between Lab Measurement & Final Rheology
Moving To Dynamic Approach… Best Operating Zone


Selecting Desired Results (X75) Reveals Pattern Of Behavior
Pattern Inherently Incorporates Process Variable Interactions
Process Camera

Pattern Of Process Variable Interactions (Red Lines) Used To
Derive Working Limits (Green Lines)… Exploited For Control
Genzyme Lessons

Leverage Standing IT Investment



Harvest New Knowledge From Existing Data





Databases
Network Infrastructure
Engage Complementary Visualization Technology
Analyze Full Span Of Process Data Available
Capitalize On Engineering Knowledge
Effectively Mine Existing Records
Exploit Gains




Problem Resolution
Response Surface Visualization
Process Optimization
Dynamic Control