Advancements in NMR Predictions- Neural Network vs. HOSE Code Algorithms Brent Lefebvre NMR Product Manager ACD/Labs’ ENC User’s Meeting April 21, 2007

New Outline Impetus behind us using NN Algorithms Implementation of NN approach Reasons why this is better than past approaches How are they accessed in the software Statistical comparisons The future of NN and HOSE code predictors

Impetus behind us using NN Algorithms

Implementation of NN approach

Reasons why this is better than past approaches

How are they accessed in the software

Statistical comparisons

The future of NN and HOSE code predictors

Why Neural Networks? The Neural Network algorithm offers a very specific advantage Speed of calculation is hundreds of times faster This enables prediction on-the-fly For Structure Elucidator, a key feature

The Neural Network algorithm offers a very specific advantage

Speed of calculation is hundreds of times faster

This enables prediction on-the-fly

For Structure Elucidator, a key feature

Why Neural Networks? Also a fresh approach for ACD/Labs to shift prediction We are always researching new ways to improve our software Also see our poster (#150) on our new increments scheme

Also a fresh approach for ACD/Labs to shift prediction

We are always researching new ways to improve our software

Also see our poster (#150) on our new increments scheme

Realization The Neural Network algorithm was outperforming our version 9 HOSE code! Steps were then taken to migrate this algorithm out of Structure Elucidator and into the ACD/CNMR Predictor

The Neural Network algorithm was outperforming our version 9 HOSE code!

Steps were then taken to migrate this algorithm out of Structure Elucidator and into the ACD/CNMR Predictor

Implementation

Neural Network Algorithm

Implementation Training the Neural Net Entire database from version 9 used Additional database of 187,000 shifts used for accuracy testing

Training the Neural Net

Entire database from version 9 used

Additional database of 187,000 shifts used for accuracy testing

Neural Network Approach How does this neural net implementation compare to others in the industry? What is unique about it? Does this make it better or worse?

How does this neural net implementation compare to others in the industry?

What is unique about it?

Does this make it better or worse?

Neural Network Approach Our research brought us to some new conclusions Some implementation details differed from previous industry attempts

Our research brought us to some new conclusions

Some implementation details differed from previous industry attempts

Neural Network Approach We found that: Characteristics of the Neural Net were NOT the most important factor Structure encoding scheme was most important Size and accuracy of training set is key Our huge quality checked database gave us a tremendous advantage

We found that:

Characteristics of the Neural Net were NOT the most important factor

Structure encoding scheme was most important

Size and accuracy of training set is key

Our huge quality checked database gave us a tremendous advantage

Using the Neural Network Predictions How are they accessed in the software?

How are they accessed in the software?

Using the Neural Network Predictions

Using the Neural Network Predictions

Limitations of the Neural Network Predictions Predictions are a black box No calculation protocol as for HOSE code Training of predictions could be possible Does not outperform HOSE code training

Predictions are a black box

No calculation protocol as for HOSE code

Training of predictions could be possible

Does not outperform HOSE code training

Statistics How do NN compare to old and new HOSE code? When should I use NN? What is the new performance?

How do NN compare to old and new HOSE code?

When should I use NN?

What is the new performance?

Prediction Accuracy We calculate our prediction accuracy for HOSE code the same way every year A “Leave-one-out” analysis of our entire database (2 million chemical shifts) This allows us to compare year on year improvement A TRUE analysis of how accurate the predictors are

We calculate our prediction accuracy for HOSE code the same way every year

A “Leave-one-out” analysis of our entire database (2 million chemical shifts)

This allows us to compare year on year improvement

A TRUE analysis of how accurate the predictors are

L-O-O Analysis Version 8.00 Version 10.05

Prediction Accuracy Standard Error of Prediction Formula:

Standard Error of Prediction Formula:

Prediction Accuracy CNMR Predictor Standard Error Version 8 - 3.11 ppm Version 9 - 2.32 ppm Version 10.00 - 2.26 ppm Version 10.05 – 1.84 ppm A 21% increase in accuracy over version 9! A 41% increase in accuracy over version 8!

CNMR Predictor Standard Error

Version 8 - 3.11 ppm

Version 9 - 2.32 ppm

Version 10.00 - 2.26 ppm

Version 10.05 – 1.84 ppm

A 21% increase in accuracy over version 9!

A 41% increase in accuracy over version 8!

Prediction Accuracy Comparison of HOSE and Neural Network >187,000 chemical shifts used in test NN algorithms- 12% accuracy increase over version 9 HOSE Code Version 10 HOSE code- 16% accuracy increase over version 9 HOSE code HOSE Code is better for now

Comparison of HOSE and Neural Network

>187,000 chemical shifts used in test

NN algorithms- 12% accuracy increase over version 9 HOSE Code

Version 10 HOSE code- 16% accuracy increase over version 9 HOSE code

HOSE Code is better for now

The Future of Neural Nets What is planned for NMR Predictors? How do Neural Networks fit into these plans?

What is planned for NMR Predictors?

How do Neural Networks fit into these plans?

The Future of Neural Nets Version 11 will further integrate the Neural Network Algorithm An intelligent hybrid approach Much like the use of incremental scheme today Stay tuned for more validation results 1 H NMR validation study

Version 11 will further integrate the Neural Network Algorithm

An intelligent hybrid approach

Much like the use of incremental scheme today

Stay tuned for more validation results

1 H NMR validation study

Acknowledgements Kirill Blinov Mikhail Kvasha Marina Solnetseva and the database team Ryan Sasaki

Kirill Blinov

Mikhail Kvasha

Marina Solnetseva and the database team

Ryan Sasaki