Sarah Nexus: Strain Information

Size: px
Start display at page:

Download "Sarah Nexus: Strain Information"

Transcription

1 Sarah Nexus: Strain Information 49 th Lhasa Limited ICGM Adrian Fowkes Senior Scientist

2 Overview 1. ICH M7 guidelines and Lhasa software 2. Sarah Nexus - Improvements to predictions Focus on structure standardisation 3. Sarah Nexus - Improvements to interpretability Compound meta data Leveraging strain information Additional compounds for review

3 ICH M7 Guidance

4 ICH M7 And Lhasa Software Database searching Data sharing Expert prediction Impurity identification ICH M7 Statistical prediction Expert assessment Test Purge mitigation Classification Archiving Reporting Control

5 Recent Updates To Sarah Nexus

6 Structure Standardisation

7 Structure Standardisation Structure standardisation is beneficial for two main reasons 1. Appropriate curation of structures to ensure the activity of compounds is accurately reflected during model building 21% of compounds with CAS numbers in the training set have at least 2 structure representations before any standardisation 2. To ensure that whatever way a query structure is drawn by the user, the same prediction is produced

8

9 Additional Information for Compounds Hypotheses Training set examples

10 Additional Information for Compounds Toggle between published and standardised structure Examine data source and follow up references

11 Additional Information for Compounds

12 Additional Compounds For Review Compounds whose activity was not resolved for inclusion into the training set are now available for review in the Nexus interface. Compounds in this panel have access to the new features in Nexus. For example, viewing strain profiles and references.

13 Using Strain Data In Sarah Models There are lots of detailed strain data available for compounds in the Sarah Nexus training set Lhasa wanted to find a way to use this strain data where it is available in the Sarah Nexus model Reduce uncertainty Better decision making Name Standardised Compounds Compound-strain pairs Vitic Nexus ISSSTY Lhasa Member => Combined

14 Options For Using Strain Data There are different ways in which these data might be used Using these data to build separate models is problematic as coverage and combination is problematic building-models-within-sarahnexuspdf/3086 Using the information to select only compounds tested comprehensively to build models is wasteful and does not take into account context data-dos-and-donts-in-building-statisticalmodels-for-ames-mutagenicity/3987 Displaying data in the correct context without changing the prediction in order to aid expert review

15 Visualisation Positive compound Activity unresolved 5 Strain negative compound Compounds where the activity could not be resolved are not included in the model training set but are available for review postprocessing Positive and negative compounds are included in the Sarah Nexus training set

16 Using Strain Data For Expert Review It is important to know if testing of negatives which are not 5 strain is sufficient The assessment on how comprehensive a compound has been tested can be done in the context of its chemical structure and potential toxicophores Expert review of strain data is particularly important in the following cases 1) Data overturns a positive hypothesis 2) Result contradicts a positive from Derek Nexus

17 Strain data in context Positive Hypothesis Negative Hypothesis Only positive examples are used to generate the strain profile Which test is most sensitive for chemicals in this class? Only negative examples are used to generate the strain profile How comprehensive is the data for supporting negative chemicals?

18 Strain profiles in Sarah Nexus

19 Strain profiles in Sarah Nexus Overall strain data for the hypothesis Strain data for the individual example

20 Interactive Dashboard Heatmap Toggle between hypotheses Access absolute figures, e.g. Positive results: 275 Negative results: 91 Toggle between supporting examples

21 Expert Review Expert review: Two relevant nearest neighbours have been tested in the most sensitive strains for aromatic amines. Increased confidence in the negative classification of these training set compounds and prediction.

22 Using Strain Data For Expert Review In most cases inspection of strain data is not required to confirm an overall call Only ~25% of overall negatives activate a positive hypothesis in available test sets 79% (3390 of 4294) of positive compounds with strain data are positive in either TA98 or TA100 94% (3965 of 4207) of negative compounds with strain data have negative data for strains TA98 and TA100 However there are specific compound classes where it may be important to test in other strains

23 Conclusions ICH M7 guidelines call for complementary expert rulebased and statistical systems for mutagenicity prediction Sarah Nexus is a transparent, well validated statistical system for use in this role Lhasa are continuously looking to improve both the performance and transparency of software to produce accurate predictions that can easily be interrogated for expert analysis Updates and improvements to training set Presentation of relevant detailed data

24 Questions? Lhasa Limited Granary Wharf House, 2 Canal Wharf Leeds, LS11 5PS +44(0) info@lhasalimited.org Registered Charity (290866) Company Registration Number