Autumn 2015 Newsletter
Welcome to the Autumn 2015 Newsletter from CompChem Solutions Limited.
In our autumnal edition we feature a personal perspective on the recent SCI/RSC Medicinal Chemistry Symposium, including an article from Al Dossetter of MedChemica Ltd on their collaborative approach to matched molecular pairs analysis and predictive ADMET which he presented on at the symposium. With the recent press focus on artificial intelligence and robots potentially replacing humans in the foreseeable future, we take a look at how this might affect us in drug discovery. With the increasing interest in computational biology across the sector, CCG have contributed an article highlighting how computational approaches can assist in the discovery of biologics, and describe an upcoming (free) course they are offering on this aspect of modelling. Do check out some upcoming events and jobs which may be of interest to our readers as well as our In Brief section for a brief round-up of news items from CompChem Solutions.
- Check out our newly published paper, in conjunction with Cancer Research Technology, describing the design and development of LIMK inhibitors.
- We’re looking forward to the UK-QSAR and Chemoinformatics Meeting on 6th October at Duxford.
- We can offer consultants for interim work or for bespoke projects, across both computational chemistry and computational biology. Contact us for more information.
18th SCI/RSC Medicinal Chemistry Symposium – A Personal Top Ten HighlightsSusan M Boyd, CompChem Solutions Ltd
I, along with what seemed like good proportion of the drug discovery community, recently attended the 18th SCI/RSC Medicinal Chemistry Symposium in Cambridge. The conference was excellent, with a variety of though-provoking talks and posters, and ample opportunity to catch up with fellow scientists during the breaks. Everyone will come away with their own personal highlights – there were so many to choose from. my own perspective, the following Top Ten highlights were particularly of interest due to their pertinence to past and present research interests.
1. Long-Range Electrostatic Effects
Alan Northrup of Merck & Co opened the conference with an interesting talk on the development of a selective SYK/Zap70 dual inhibitor for the treatment of rheumatoid arthritis. One of the key factors in the design of the key compounds was a long-range electrostatic effects. These are field-based effects which are non-directional, and can be significant across a broad range of distances. Compounds experiencing a long-range electrostatic effect will show SAR requiring a particular electrostatic property within a region of the molecule, but not at a specific substituent position in the structure. It is quite unusual to find application of these effects in drug discovery, but in this instance the effect was crucial to the design of potent molecules.
2. Targeting Disease-Linked Mutations in Proteins
Several talks (Robert Heald of Argenta on the EGFR kinase domain and Jason McCartney of Vertex on the Cystuc Fibrosis Transmembrane Regulator) described the targeting of proteins which contain particular disease-linked mutations. These approaches are likely to become a key area for the future as personalised medicine becomes more widespread. The kinase inhibitor developed is a covalent binder, which links to the gatekeeper Cys in a carcinoma-mutated form of the protein whilst the cystic fibrosis treatment developed arose from HTS followed by full drug discovery effects, resulting in a product which was approved in 2012 for patients with specific mutations in the CFTR gene.
3. Gini Scores in Kinases
Sarah Skerratt of Pfizer in Cambridge (UK) gave a talk on the discovery of Trk inhibitors. The approach taken made extensive use of Gini scores for the binding site residues targeted when designing the structures. The Gini Scores highlight residues in the site likely to be key for selectivity. For example, gatekeeper residues and back-pocket residues typically have higher Gini scores than hinge residues. The principle behind design is to target greater ligand interaction with the key residues by Gini score.
4. Proteolysis-Targeting Chimeras (Protacs)
Ian Churcher of GSK talked about the use of Protacs which are small molecules designed to bring a target protein into close proximity of a ubiquitin E3 ligase which promotes ubiquitination and further degradation of the target protein. This new approach has the potential to impact therapeutic targets in a way not currently possible withmore standard small molecules.
5. Protein-Protein Interaction (PPI) Rings Analysis
Richard Taylor of UCB presented on the subject of targeting protein-protein interactions using small molecule fragments, and described their recent ring analysis work in which ring fragments found in the FDA’s orange book of drugs are distributed across protein target families. In total only 1197 ring frameworks and 351 unique rings were identified in the study. A case study was presented where an SPR fragment screen was conducted across a range of PPI targets, where 525 unique ring framework systems and 194 unique rings were found to bind to PPIs.
6. ADMET Solutions Using Big Data
Al Dossetter of MedChemica Ltd presented on this topic. See Al’s article for more information!
7. Nav1.7 Inhibition
Nigel Swain of Pfizer in Cambridge (UK) gave a fascinating talk on discovery of selective inhibitors of the voltage-gated ion channel Nav1.7. The inhibitors target a novel site sitting towards the outside of the channel, unlike most of the known inhibitors which target the inactive state.
8. Imaging Mutant Huntingtin Aggregates: Development of Potential PET Ligands
Mike Prime of Evotec (UK) Ltd gave a nice talk on the development of potential ligands for PET imaging of Huntington’s Disease which they are exploring with CHDI. In the absence of in vivo biomarkers for mutant Huntingtin (mHTT) Successful PET imaging could allow observation of mHTT aggregates in the brain of patents with Huntington’s Disease. Compounds developed so far show selective regional binding to diseased over wild-type brain samples.
9. Fragment-Based Discovery of a Novel Binding Pocket in Lipoprotein-Associated Phospholipase A2 (Lp-PLA2)
Alison Woolford of Astex described a project conducted in collaboration with GSK to use FBDD to discover inhibitors of Lp-PLA2. The fragment screen identified four fragments which all bound at complementary positions in the site, including one which induced a reside flip to generate a novel pocket in the site. Further development has led to a novel potent lead structure with improved physicochemical properties over the known Lp-PLA2 inhibitors.
10. Risk/Decision Making in Drug Discovery
Steve Swanson of GSK presented a highly entertaining piece on thinking skills, drawing on his personal experience of playing chess at international level for Scotland and exploring how we might make better decisions on risk in drug discovery. He described “system 1” and “system 2” thinking, whereby system 1 is characterised by a more “autopilot” mode of thinking, whereas system 2 involves deeper consideration of the task. He enlightened us as to the 7 Deadly Chess Sins (thinking [in too much depth], blinking, wanting, materialism, egoism, perfectionism and looseness) and concluded with the sentiment that if you think you’ve got the right answer – think some more!
Will Robots Nab Our Jobs?
Susan M Boyd, CompChem Solutions Ltd
With the recent publicity surrounding artificial intelligence, I thought it might be interesting to explore how pharmaceutical and other drug discovery jobs might fare in the current predictions. The BBC website features an interactive risk calculator which uses data adapted from an original study performed by Oxford academics to derive a percentage score of the likelihood of particular job functions becoming fully automated over the next 20 years. Of course, as scientists used to deriving and working with scoring functions and statistics, we could discuss at endless length (I’m sure!) the quality, validity and utility of the results – but that is a whole ‘nother ball game and a topic for another time perhaps. Taking the results of the analysis at face value, it does seem to bear out what we have long suspected. Finance officers are eminently replaceable (97% likelihood of automation) but creative, highly trained scientists are not (chemical scientists have 6% likelihood of automation)! There were, however, a number of surprises in the study. HR workers, for example, have a 90% risk of automation and Health and Safety Officers a 53% risk. But the key trends are generally quite clear and in line with expectations. The more innovative thought or people skills are required for our jobs, the harder it will be to replace us. Academics are sitting very pretty in this study, with only a 1% risk of automation. Lab technicians, however, are much more at risk with a 70% chance of automation. One interesting observation is that CEOs are a little more likely (9%) than chemical scientists (6%) to be replaced by automatons! Presumably scientifically trained CEOs who run highly technical or scientific businesses are a little safer than that.
In our line of work many organisations are increasingly encouraging non-computational scientists to run computational chemistry calculations themselves. Of course this is, in itself, a form of automation of our role. However, as long as the computational scientists have input into the risks and benefits of particular applications being rolled out to the wider scientific community of our organisations, and are themselves still driving the science forward in ways which can benefit future research, we should still have a critical role to play in the overall scheme of things. The bottom line is still “innovate to survive” so let’s keep on pushing those boundaries!
Roche and AstraZeneca share R&D ‘Knowledge’ to Accelerate Drug Discovery
Medicinal chemistry data-sharing consortium makes R&D knowledge available to global researchers in charities, universities and research foundations through MedChemica and Elixir Software’s SaltTraX and consultancy.
Al Dossetter, MedChemica Ltd
As part of their goal to accelerate the discovery compounds with a higher probability of clinical success, the collaboration announced by Roche and AstraZeneca has delivered a database of chemical modifications to address pharmacokinetic and toxicity issues. This database is founded on using a dedicated technology (Matched Molecular Pair Analysis, MMPA) which enables the sharing of pre-competitive knowledge without divulging confidential chemical structure information.
At the outset, the companies committed to making the data generated available to the broader research community, particularly research foundations, charities, academia and contract research organizations that serve these groups. Some thirteen organizations have joined the initiative to date, and are trialing the drug design online platform based on the database. Early adopters of the knowledge are Cancer Research UK (CRUK – Manchester research site), Institute of Cancer Research (ICR), Structural Genomics Consortium (SGC Toronto research site), Universities of Leeds, Warwick, Sheffield, Dundee, CapeTown, and Unicamp (São Paulo), Galapagos, C4X Discovery, Sygnature, Medicines Malaria Venture.
Data-sharing is managed by MedChemica Limited who apply their expertise in the key technologies of MMPA, Big Data platforms and statistical analysis. MMPA enables the sharing of knowledge as only fragments of molecules are shared between the contributors thus removing any concerns about compromising intellectual property. Statistical analysis of the combined output selects ‘design rules’; transformations from one chemical group into another that has improved ADMET properties for the contributors compounds in the past. These rules are supplied back to the contributors as the ‘grand rule database’ of knowledge. A further collaboration between MedChemica and Elixir Software Limited provides the secure, online design tool SaltTraX© and though this makes the knowledge available for rapid use by research foundations, charities, academia and the life science bio-tech industry. The tool is available to use on a pay-per-query or annual license fee model.
The large pharma data-sharing consortium is open for other large companies to exploit the available resource through additional added knowledge. The more data added to this system raises the quality and specificity of drug design rules, returning a benefit to all participants of efficiently applying the most useful medicinal chemistry know-how to identify potential new drug candidates, and using fewer rounds of design, synthesis and testing.
The whole life sciences industry is under pressure to deliver value to returns to patients, investors, donators and public funding alike. Knowledge sharing fulfills the need to speed up and produce more for less and help treat a broader spectrum of diseases by enabling this process. The knowledge base is particularly powerful for blue sky researchers in small organizations as the database is extensive and can suggest new molecules to make based on hard won knowledge of years of effort in large pharma. There is a great deal yet to learn and many savings can be made with this knowledge and collaborative science that enriches the life science eco-system and brings real benefit to all.
Modelling of Biotherapeutics
Andrew Henry, Chemical Computing Group Inc
There are two common areas for molecular modeling in the development of antibodies as biotherapeutics. In the first, an antibody will have been identified with interesting binding properties, but it may have problems such as low affinity, low solubility, aggregation, and poor stability. Here the aim is to make a small number of changes to improve those properties, without making major changes that might affect the overall binding of the antibody to the target of interest, and off-target selectivity. The second is at an earlier stage, when amino acid sequences have been determined for a set of different antibodies. Here the aim is to use modeling to highlight issues that might arise during the development of the antibody. This requires high throughput methods, but the aim is to act as a filter to ignore antibodies with potential issues, rather than trying to correct them. Some properties might be identified from the sequence alone, but for others, building a homology model of the antibody can help to describe the characteristics of the molecule, such as the arrangement of charged and hydrophobic residues in the structure. It is useful to annotate these patches onto the sequence and the structure, and then to compare the residues that make up these patches, to see if they are conserved in other antibodies.
Natural IgG antibodies are formed from two heavy and two light chains, arranged in a number of folded domains which are joined by flexible linker regions. Structures have been determined using X-ray crystallography for thousands of antibody Fab fragments, which includes the Complementarity Determining Regions (CDRs) of the antigen binding site. These domains have a highly conserved core structure, and are relatively rigid. The CDRs vary in sequence and in structure, and have been found to be more flexible using NMR and X-ray crystallography. There are only a few X-ray structures of whole antibody molecules, as it is difficult to crystallise proteins where the domains are joined by flexible linkers. Homology modeling is the most accurate for rigid domains, which are found in a lot of template structures. For flexible loops and linkers an ensemble of models may be required to describe the set of possible structures.
One of the major issues in bringing a therapeutic antibody to market is the cost of synthesising the protein, so that it can be delivered at high concentration, to deliver a high dose in a small injection of liquid. At high concentration, the protein might form aggregates which can be toxic. Large hydrophobic patches on a protein surface have been identified as a common motif in antibodies which are prone to aggregation1.
At low concentrations, an overall charge on the protein will decrease the aggregation and viscosity of the solution, by electrostatic repulsion. However, in higher concentration solutions the crowding of neighbouring molecules reduces the impact of the electrostatic repulsion. This was shown in an example from Pfizer, where neutralization of a negatively-charged surface patch reduced the viscosity of antibody2.
Glycosylation on the surface of a protein can make it more difficult to prepare consistent samples of the protein during manufacturing. It is useful to predict potential glycosylation sites. Another issue is the stability of the antibody molecules, to limit the changes in the structure due to deamidation of Asparagines, isomerization of Aspartates, and oxidation of Methionine and Tryptophan residues3.
Asparagine and Aspartate degradation was found in 15 of 37 of a set of therapeutic antibodies. All of these degradation sites were in the CDRs4. The most important descriptor in a decision tree to classify the reactivity of Asn and Asp residues was the flexibility of the structure in a set of homology models.
MOE is a molecular modeling package which can build and analyse homology models of antibodies. On September 30th, there will be a free training course in Cambridge for biologics modeling with MOE.
Biotherapeutics Modelling Course, free course (registration essential) run by Chemical Computing Group, 30th September 2015, Cambridge, UK.
Cambridge New Therapeutics Forum, 13th October 2015, Cambridge, UK
UK Drug Discovery Centres Meeting, 26-27th October 2015, London, UK
Cambridge Chemoinformatics Network Meeting, 26 November 2015, Cambridge, UK
MGMS Young Modellers Forum, 27th November 2015, London, UK
Joint PCF/UK QSAR & Chemoinformatics Group Spring 2016 Meeting, 15th-16th March 2016, GSK Stevenage
Computational Chemistry / Computer-Aided Drug Design, Heptares, Herts, UK
Senior Software Developer, Chemical Computing Group Inc, Cambridge, UK
Application Scientist, Chemical Computing Group, Cambridge, UK, or Cologne, Germany
Programmer, Cresset, Cambridge, UK
Senior Computational Chemist, Charles River Laboratories, Cambridge, UK
Computational Chemist, Drug Discovery Unit, University of Dundee, Scotland
Molecular Modeller, Astex Pharmaceuticals, Cambridge, UK