ISPE 2019 Driving Step Changes in Manufacturing Operations with Predictive Insights

Driving Step Changes in
Manufacturing Operations with
Predictive Insights
Practical Application of Pharma 4.0

2019 ANNUAL MEETING & EXPO
27-30 October | Las Vegas
CSO & co-founder of Bigfinite
Physics Degree, Master in Information and Knowledge Society, Post graduated in quality systems for
manufacturing and research pharmaceutical processes.
7 US Patents: encryption, transmission, storage and processing big data for regulated environments in the cloud
Articles & white papers based on cloud, pharmaceutical industry and data science
AI Health Xavier University of Cincinnati: AI Core Team and AI Manufacturing Team Lead
PDA. Scientific Committee & Europe co-chair
AI & big data SME for Sciences in the Spanish Parlament
Professor at the University Autonomous of Barcelona
Bioinformatics of Barcelona. Project Leader of the Data Integrity team
Toni
Manzano

Pharma 4.0 ≈ Digitization + ICH Q10

What does big data mean from a regulatory perspective?

What does big data mean in GxP environments?
3600000 GB = 3600 TB per minute
Data transferred now

Yet the adoption rate of big data, cloud technologies in Pharma is
lagging that of other industries

Companies investing in AI by industry

Opportunities enabled by digital and analytics are recognized
across sectors
Digital Quotient score. Points (out of 100)

Pharma 4.0. Is this strategy well aligned?
2017: A Holistic Approach to Production Control: From Industry 4.0 to Pharma 4.0.
Herwig C., Wölbeling C., Zimmer T.
2018: Getting Ready for Pharma 4.0
Manzano T., Langer G.
2019: Pharma 4.0 – The New Frontier for the Pharma Industry
Minero T., Augeri A.
2019: The ISPE Pharma 4.0 Operating Models
Heesakkers H., Schmitz S., Kuchenbrod U., Wölbeling C., Zimmer T.

Something is changing the world perspective regarding AI

Something is changing in Pharma regarding AI

Area Priority
"Big Data" Leverage "Big Data" for regulatory decision-making
Biocompatibility Modernize biocompatibility and biological risk evaluation of device materials
Real-world evidence Leverage real-world evidence and employ evidence synthesis across multiple domains in regulatory decision-making
Clinical performance Advance tests and methods for predicting and monitoring medical device clinical performance
Clinical trial design Develop methods and tools to improve and streamline clinical trial design
Computational modeling Develop computational modeling technologies to support regulatory decision-making
Digital Health and
cybersecurity
Enhance the performance of Digital Health and medical device cybersecurity
Healthcare-associated
infections
Reduce healthcare associated infections by better understanding the effectiveness of antimicrobials, sterilization and
reprocessing of medical devices
Patient input Collect and use patient input in regulatory decision-making
Precision medicine and
biomarkers
Leverage precision medicine and biomarkers for predicting medical device performance, disease diagnosis, and
progression
CDRH Regulatory Science Priorities
CDRH's regulatory science priorities serve as a catalyst to improve the safety, effectiveness, performance, and
quality of medical devices and radiation-emitting products, and to facilitate introducing innovative medical
devices into the marketplace. Report of August 22th, 2019

What is Artificial Intelligence?
“ AI can be thought of as simulating the capacity for
abstract, creative, deductive thought - and particularly the
ability to learn - using the digital, binary logic of computers ”
Artificial Intelligence (AI) is no longer some bleeding
technology that is hyped by its proponents and mistrusted by
the mainstream. In the 21st century, AI is not necessarily
amazing. Rather, it is often routine. Evidence for the routine
and dependable nature of AI technology is everywhere.
“Verification and Validation and Artificial Intelligence”, Tim Menzies, Portland
State University, Charles Pecheur, NASA Ames Research Center. July 2004

Data must be prepared before to use it:
DS invest the 80% of their time

Warning Letters including references to
data management and data integrity
Source: fda.gov & PharmaceuticalOnline

Expected and required data quality

When the statistical results have huge impact in the users

More than just multivariable models...

5 AI concepts
● Training/Test data
○ Problem’s dataset
● Algorithm
○ Mathematical procedure that creates the Model from the training data
● Model
○ Mathematical system that has been created from the exploration of a data set. It’s created
after an extensive learning process referred as training.
● Prediction / Classification / Recommendation / Recognition
○ Single inference over a model with an unseen sample
● Evaluation
○ Score evaluation of the Test dataset

Supervised and unsupervised algorithms
Learning Supervised Learning UnSupervised
➔ Linear Regressor
➔ Support Vector Machines
➔ Random Forests
➔ Neural Networks
➔ K-Nearest Neighbours
➔ Gradient Boosted Trees
➔ ...
vs
➔ K-Means Clustering
➔ Hierarchical Clustering
➔ Isolation Forests
➔ Graphical Lasso
➔ Bayesian Networks
➔ Markov Hidden Models
➔ ...
AI unsupervised Vision ExampleAI supervised Vision Example

Supervised Learning. Decision Trees.
From a business decision point of view,
a decision tree is the minimum number
of yes/no questions that one has to ask,
to assess the probability of making a
correct decision, most of the time. As a
method, it allows you to approach the
problem in a structured and systematic
way to arrive at a logical conclusion.
Usually applied for root cause analysis

Supervised Learning. Naïve Bayes Classication.
• P(A|B) is posterior probability
• P(B|A) is likelihood
• P(A) is class prior probability
• P(B) is predictor prior probability
Naïve Bayes classifiers are a family of simple
probabilistic classifiers based on applying
Bayes’ theorem with strong (naïve)
independence assumptions between the
features.
Some examples of usage:
• Deviations classification
• Clinical trials analysis
• Cause-effect analysis

Supervised Learning. Naïve Bayes Classification.

Supervised Learning. Logistic Regression.
Is a way of modeling a binomial outcome with
one or more explanatory variables. It measures
the relationship between the categorical
dependent variable and one or more
independent variables by estimating probabilities
using a logistic function, which is the cumulative
logistic distribution.
In general, regressions can be used in pharma applications such as:
• Predicting the quality of batches (ok / ko)
• Unexpected manufacturing stops
• Predicting the time to finish of a certain product
• Is there going to be a non conformity in a batch?

Unsupervised Learning*. Principal Component Analysis
PCA is a statistical procedure that uses an orthogonal transformation to convert a set of observations of
possibly correlated variables into a set of values of linearly uncorrelated variables called principal
components. It is not suitable in cases where data is noisy (all the components of PCA have quite a
high variance). Notice that domain knowledge is very important while choosing whether to go forward
with PCA or not.
Some of the PCA applications include:
• Compression
• Simplifying data for easier learning
• Dimensional reduction
• Identification of relevant dimensions
• Dimensionless analysis
* PCA can not be consiered as an AI algorithm

Which types of software must be validated?
• All medical devices that contain software – even Class I devices – are subject to the Design
Control Provisions, and thus require software validation
• Any software used to automate any part of the device production process or any part of the
quality system must be validated for its intended use (21 CFR §820.70(i)).
• Computer systems used to create, modify, and maintain electronic records and to manage
electronic signatures (21 CFR §11.10(a)) must be validated.

Not adequately validating software can lead to a Warning Letter...
4/12/2017
3. Failure to ensure that design verification shall confirm that the design output meets the
design input requirements, as required by 21 CFR 820.30(f). For example: Your firm has a
design input, (b)(4), of “the Remote Monitoring device shall only open network ports to
authorized interfaces” which is documented in Software System Requirements
Specification, Document. This is implemented as a design output in your firm’s DeviceName
Software Requirements Specification Uploads (b)(4).
This design output was not fully verified during your firm’s design verification activities.
According to your firm’s testing procedures, (b)(4), Final Configuration Test Procedures,
(b)(4) and Final Configuration Test Procedures Document (b)(4), the requirement was only
partially verified by testing that the network ports opened with an authorized interface. Your
testing procedures did not require full verification to ensure the network ports would not
open with an unauthorized interface.

Trust in AI
Robustness Fairness Explainability Lineage
Ex. Adversarial
training methods
Bias
• Data collection
• Processing
• Labelling
Explainer
•Data
•Metadata
•Catalog
•Hyperparameters
•Governance
•Processing
•Integration
•Privacy.

Trust in AI
Adversarial training methods example:

Use cases in Pharma: Root Cause Analysis
?
CQA: All of them under specs
CPP: Equivalent results except culture
duration

Getting knowledge from AI in primary packaging
LC BCO Set-up Warm-up WO Reworks Reconcilliation
•Pens / minute
•Product / format
•Batch size
•Day of the year
•Time of the day
•Shift
•Downtime
•Stops
•Affected units by stops
Theoretical OEE Realistic OEE Deep process knowledge

Use cases in Pharma: Unpurified Bulk bag selection
end
Start
unpurified bulk
manufacturing
drug substance
purification
From cell bank
To CMO for
formulation and filling

Use cases in Pharma: Anomaly detection
Predict TTNF (Time To Next Failure)
Recommend actions to avoid Next Failure

Section Divider
Slide Layout
Section divider subhead

ISPE 2019 Driving Step Changes in Manufacturing Operations with Predictive Insights

Recommended

Recommended

More Related Content

Similar to ISPE 2019 Driving Step Changes in Manufacturing Operations with Predictive Insights

Similar to ISPE 2019 Driving Step Changes in Manufacturing Operations with Predictive Insights (20)

Recently uploaded

Recently uploaded (20)

ISPE 2019 Driving Step Changes in Manufacturing Operations with Predictive Insights