Towards Efficiency in Uncovering all Potential Longevity-Altering Substances

The research community is moving, slowly and incrementally, towards a world in which drug libraries become vastly larger and more useful because it should be possible to use computational techniques to far more efficiently (a) predict the effects of specific compounds on specific biological mechanisms, and (b) design similar, better compounds. Much of the trial and error, and thus most of the cost of drug discovery will go away. The result will be a pharmaceutical development processes that is still definitely of a trial and error nature at its core, but much more informed, far removed from the blind fumbling and chance discovery of the past. Insilico Medicine is a business community example of progress towards this goal, and the open access paper noted here is an example of analogous research community work.

Does this mean we should expect the near-term emergence of longevity-enhancing drugs based on the adjustment of metabolic state that are vastly more effective than rapamycin analogs? I think no. My thesis is that the effectiveness of these drugs is far more constrained by the lack of plasticity of human longevity in response to metabolic alteration than it is by the quality of the drug. Exercise and calorie restriction mimetics have a limited upside in terms of what they can do for us.

Where pharmaceutical approaches do prove to have larger and more reliable effects on longevity, it will be because they are producing true repair of the causes of aging rather than mere metabolic adjustment. Examples include removal of senescent cells, or breaking down forms of metabolic waste ranging from cross-links to amyloids to the constituents of lipofuscin that accumulate in lysosomes. Here, better drug development processes will lead to an improved pipeline of drug candidates that are more efficient at specific damage repair tasks. This is where upgrades in the infrastructure of the traditional pharmaceutical pipeline can shine. Now if only more groups were intent on this path rather than trying to find a marginally better alternative to rapamycin...

Old age is the greatest risk factor for many diseases, including various types of cancer, inflammatory and neurodegenerative diseases. Traditional medical science combats one disease at a time, instead of combating the underlying biological ageing process that leads to many age-related diseases. From a whole body system's point of view, this traditional one-disease-at-a-time approach focuses on the downstream diseases, rather than considering the underlying mechanisms of age-related functional decline. This approach has limited effectiveness at present and is likely to be less effective in the future, because of an increasingly larger elderly population suffering from multiple age-related diseases. In contrast, interventions that slow down ageing and promote "healthy ageing" could in principle delay the onset of all age-related diseases, with a significant benefit to human health and a large reduction of healthcare costs.

Pharmacological interventions are arguably the most practical ageing intervention for humans, avoiding the main problems with genetic interventions (generally unethical in humans) and dietary interventions such as caloric restriction, which are difficult to maintain for the vast majority of people. For instance, there is currently great interest in discovering drugs that mimic the process of caloric restriction. In addition, promising research on pharmacological interventions on the ageing process is underway at the National Institute of Aging's Intervention Testing Program (ITP), which consists of administering drugs or chemical compounds to mice under carefully controlled conditions. However, as mouse experiments are costly and time consuming, so far only a limited number of drugs or compounds have been evaluated. Thus, using simpler model organisms for evaluating a chemical compound's effect on an organism's lifespan is appealing, and a substantially larger number of studies have administered compounds to C. elegans than other organisms. As the ITP for mice, the Caenorhabditis Intervention Testing Program has been introduced for assessing longevity variation for chemical compounds.

In this work we analyse data from the DrugAge database, which contains information about chemical compounds and their effect on the lifespan of organisms. DrugAge contains a variety of compounds with anti-ageing properties such as gerosuppressant, geroprotective and senolytic activity as well lifespan increasing properties for a specific species. In order to analyse such data, we use random forests, which is a supervised machine learning method. In this work, the random forest builds a classification model to predict whether or not a chemical compound will increase the lifespan of C. elegans, based on predictive features describing that compound. The best model produced by the random forest method was applied to a screening "external" dataset with compounds from the DGIdb database, where the effect of the compounds on an organism's lifespan is unknown. The predictions of that model were used to identify the "top hit" compounds in the DGIdb dataset, i.e. compounds with higher probabilities of increasing lifespan in C. elegans.

In conclusion we have built, using machine learning, a model to predict the longevity effects of chemical compounds in C. elegans, using the recently published DrugAge dataset. The list of top-hit compounds and their analysis contributes to our knowledge of likely longevity-extending compounds, and experimental confirmation of these predictions would be an interesting direction for future research.



Hmmm, the whole thing has a whiff of 'systems biology' about it to me. I don't think you can simulate the complexity of an in vivo model, at least not in the near term.

Still, I'm just a layperson ranting on the internet. And one very good thing is that the founders of Insilico are positive about the possibility of doing something about aging, and are obviously well connected to funding sources.

Even if Insilico's algorithms do not lead to a reduction in drug development costs, there seem to be other areas of medicine that are in need of algorithmic prediction. For example in the recent phase 1 test shake out of cancer neo-antigen vaccines, algorithms were used to predict which neo-antigens would bind best with T cell receptors:

"Although neoantigens were long envisioned as optimal targets for an antitumor immune response, their systematic discovery and evaluation only became feasible with the recent availability of massively parallel sequencing for detection of all coding mutations within tumors, and of machine learning approaches to reliably predict those mutated peptides with high-affinity binding of autologous human leukocyte antigen (HLA) molecules."

Posted by: Jim at August 17th, 2017 1:23 PM

These computational drug discovery approaches have been around the pharma system for years in different guises - they've yeilded zilch and do nothing to shorten the real time frame of drug development - but they do pop up in cycles every decades of so and siphon less savvy investors money away

Posted by: Chaim Lezonno at August 17th, 2017 4:49 PM

Post a comment; thoughtful, considered opinions are valued. Comments incorporating ad hominem attacks, advertising, and other forms of inappropriate behavior are likely to be deleted.

Note that there is a comment feed for those who like to keep up with conversations.