[MUSIC]

So welcome to section five of this lecture.

We have now seen how a database has been created,

which is very promising for data mining.

And obviously, is also very promising now to carry out Read-Across.

And in order to make the use of this tool easy as possible,

we're going to discuss now an automated tool, which in its making.

So it's a work in progress which we're presenting, but

is also something which is demonstrating some of the challenges on the road.

Let's come to the last element, which is help tools.

What Read-Across cannot be done well manually because you have a lot of

fingerprinting, you have a lot of comparisons.

And ideally, you're comparing to more than one chemical in order to do Read-Across.

This is showing you not even exhaustive list of tools for Read-Across.

They have different goal and they're helping with different type of elements.

This is only showing you that what I'm going to present next is not alone.

There is tools available.

Many of them are free.

Some others require either subscription or buying access to respective databases.

Our vision was now, and this is described in this article, which was the same

journal of ethics as the previous articles on good Read-Across practice and

the database.

This Food for Thought article I'm describing our vision for

creating a tool which is automated.

Which is producing Read-Across based on the similarity space around a chemical.

So the idea is that event based tool would allow to enter a structure,

and possible also proprietary data the person has.

We would then make a prediction from the nearest neighbors of this chemical.

And we would use the bit larger region to assess the certainty of our tools.

And then we would like to produce something which is helping with

registrations.

So REACH-concordant reporting which means in this case IUCLID 5, the software,

the mean time upgrade it to IUCLID 6 a few weeks ago would in a copy paste,

take the results of this type of Read-Across.

So to show the process, this is the chemical universe.

With the question mark in the center, we are showing a substance for

which we are lacking data.

We would use the local similarity space to make a prediction.

So we would say, based on the most similar substances,

what is most likely the property of this substance?

And you would use the region to express how certain we can be.

Is this a part of the chemical universe where there's a lot of activity cliffs,

where there's a lot of change in properties?

Then we have to simply assign a higher uncertainty than when there is

homogeneous behavior with respect to this hazard.

So this is the screenshot of a better version of the tool under development.

We're just showing you here an example for lactic acid.

Lactic acid is obviously a pretty harmless substance, so we're not surprised that

we don't see, this is the pie chart to the left, which is showing you this skin

sensitizers in its surrounding, that everything is non skin sensitizing.

However, if you would change and click to the pie chart to the very right,

there is eye irritation observed in the neighborhood of the substance.

And this is not surprising because it's an acid.

And an acid in the eye will lead to eye irritation.

Another example, this is a substance which is a very non-skin sensitizer,

and as you can see, there's quite a bit of dark spots around the substance.

And these dark spots indicate there's skins sensitization in the neighborhood of

this substance.

So there's a lot of prospects at the moment.

New features are being implemented an ability to actually predict

new substances.

So even before a chemical is synthesized it could be

evaluated whether it has likely certain properties.

We are exploring different alternative similarity measures going beyond

[INAUDIBLE].

We are introducing filters, for

example you might have heard about Klimisch score which is a quality score,

which is assigned to say this was a high quality study or not.

So we can restrict our predictions on high quality studies.

We can use different prediction models.

We can include and incorporate additional databases from other sources.

The optional inclusion of proprietary data owned by the person interested in

the assessment, but also direct comparison one on one in order to identify is

this a substance which has better properties?

Is this, in terms of chemistry, you might consider for using instead?

Or opportunities to work on a list of substances.

So imagine a long list of substances which are on the products of a given company,

they will simply like to know what other ones we should look at,

which other ones of concern.

And running such a list of substances is one of the features currently implemented.

And then reporting compatible with IUCLID of the European Chemical Agency

to make it easy to use this type of data gap filling for the registration

process with the European Chemical Agency's to cells data in that line.

However, in the end, the bitter truth, or the litmus test for

all essays of this kind, is validation.

So we're discussing already with the National Institute of

Environmental Health Sciences and the Food and Drug Administration in the US,

the path towards validating this tool.

And we're preparing for a conference on read-across acceptability

by regulators in March, 2017 here in Baltimore.

This is opening up for

opportunities, especially also in the area of green chemistry.

We are interested, especially, in the aspects of toxicity here.

Because two of the principle of green chemistry,

which is sustainability initiative in chemistry, benign by design,

which means chemists produce things which are likely not to have toxic properties.

Or reducing toxicity by eliminating toxic substances early on in product

development process by testing early and choosing substances of lesser toxicity.

So this front loading of toxicity assessments lends itself to read-across

methodology because they are fast, they are cheap, and

can produce information at a early stage.

Which is not perhaps is not of the necessary quality of the regulatory

registration, but

which is informing what might be a better choice when everything is still open.

So as a kind of summary, the question might be posed,

is this the now the next hype because it is such a young development?

Read-across is really coming up over the last decade, or is it a game-changer?

I tend to say it is a game-changer.

I am impressed by the fact that we are only using local validity.

We're not trying as many other computational methods to find the formula

for the entire chemical universe to predict and calculate a property.

They're just looking what do we know about substances which

are very similar just around.

For me it was quite impressive to see the power of big data.

These 10,000 substances are forming already 1.5 million pairs.

And this is increasing rapidly with more substances being added.

And the more pairs, the more information we have on each and every substance.

There's also now the interesting cross talk between the chemical structure and

biological support data, which promises to help us to get around

activity cliffs which we would not observe based on structure only.

There's some changes, and for the next level it will be important to combine

the generation of the read-across with the read-across assessment framework.

And other ways of evaluating the results of read-across or

what the European Chemical Agency has put forward is certainly pioneering here.

And we need to measure what we deliver against these frameworks.

We will need more data. We need data which are of high quality and

well created.

We need to work on standardizing the biological data.

At the moment, it's very few substances which have been tested in standardized

protocols which are publicly available.

But this is for the years to come.

I hope I have given you an overview on an exciting approach on technology which is,

at the moment, coming into toxicology, being more broadly used.

And is making information available, sometimes within seconds, for

which traditional assessments might take several years.

As always, I would like to close with a quote.

In this case, it's a pretty old one, it's from Aristotle.

And he said, it is the mark of an instructed mind to rest easy with

the degree of precision which the nature of the subject permits.

And not to seek an exactness where only an approximation of the truth is possible.

And in this sense, thank you very much for listening.

[MUSIC]

Creating an Automated Tool for Read-Across

Toxicology 21: Scientific Applications

Conheça os instrutores