Replies to post #123787 on Anavex Life Sciences Corp (AVXL)

[Fireman02360]

Crickets.......

[kld2]

KEM(R) – or Knowledge Extraction and Management – is an association rules-based (non-statistical) analytical technology that finds patient responder sub populations and biomarker signatures that statistical methods are unable to detect. KEM optimizes clinical trial inclusion/exclusion criteria thereby lowering the number of patients needed to reach clinical endpoints and saving sponsors both time and money.

Next month's PK/PD oral communication sessions should be interesting. I hope we are not disappointed.

[OFP]

Ariana Pharma, a provider of “non-statistical” clinical data analysis technology

You don't see how this can be true and yet they still provide statistical analysis?

combining heterogenous data from multiple sources with patient characteristics

You don't see that when the data is numerical this is, by definition, a statistical exercise?

The problem is that the KEM hierarchical clustering method as used for AVXL is just a way to organize data process to predict links between baseline data and efficacy outcomes. This use of a sample of numerical data to make general predictions of relations across the whole set is in itself is a statistical process

However, there is more to the Ariana process. Next they will only allow that those relationships to be called "predictors" after they test the validity of those relationships using (gasp) standard statistical measures (P<0.05 based on Fisher exact tests)

Finally, they express efficacy relative to placebo in odds ratio...another standard statistical measure.


Didn't see any confidence intervals but did see p-values and its alpha which imply null and hypothesis. There were % signs aplenty.

So after all the wasted words above you could just say finding correlations within numerical data that can be applied to the whole group is a statistical exercise (analysis). How they get to the correlations doesn't change that fact.

Finally, no you can't hide behind the fact that KEM is rule based or that SOMETIMES certain ASPECTS of Ariana's analysis falls outside of or on the margin of statistics. Nor can you blame them for trying to distinguish their product in the market by emphasizing the non-statistical elements. Clearly their comprehensive analysis, when assessing drug effects as is the case with AVXL, is a statistical analysis.

[bas2020]

Well played, Mycroft!

[MycroftHolmes]

Re: KEM Process Internals

For those interested in such things, here is a document that explains the internals of the KEM process.

As previously stated, KEM seems to be a rules based topological association engine.

The paper can be downloaded for free by hitting "download" here:
Novel Rule-Based Method for Multi-Parametric Multi-Objective Decision Support in Lead Optimization Using KEM

We introduced the methodology developed in KEM which presents the advantage of building an ontology from complex data. This process derives knowledge in an exhaustive manner, learning from the entire dataset, including outliers and non-statistically significant data. The ontology described as a set of simple rules, handling both presence and absence of properties, is used for extracting SAR information, for multiobjective optimization and for endpoint prediction.

One can expect that more novel applications which exploit the advantages of machine learning methods will be published in the coming years, with emphasis on decision support in drug discovery. Exploiting the knowledge acquired by many diverse projects along the years, and capturing unique, statistically non significant, information-rich events constitute one of the biggest challenges for optimizing the drug discovery process.