{"id":13091,"date":"2025-09-15T21:02:25","date_gmt":"2025-09-15T20:02:25","guid":{"rendered":"https:\/\/www.blopig.com\/blog\/?p=13091"},"modified":"2025-09-15T21:02:27","modified_gmt":"2025-09-15T20:02:27","slug":"reflections-on-grc-cadd-2025-a-week-of-insight-innovation-and-baseball","status":"publish","type":"post","link":"https:\/\/www.blopig.com\/blog\/2025\/09\/reflections-on-grc-cadd-2025-a-week-of-insight-innovation-and-baseball\/","title":{"rendered":"Reflections on GRC CADD 2025: A Week of Insight, Innovation, and Baseball"},"content":{"rendered":"\n<h2 class=\"wp-block-heading\"><strong>Henry<\/strong><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Back in July, some very lucky OPIGlets ventured across the pond to discover life in Southern Maine (and Boston!). For someone visiting Boston for the first time, no trip would be complete without a Red Sox game\u2014a thoroughly enjoyable highlight (see Figure 1). While we were there, we also went to Gordon Research Conference (GRC) on Computer Aided Drug Design (CADD).<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full\"><a href=\"https:\/\/i0.wp.com\/www.blopig.com\/blog\/wp-content\/uploads\/2025\/09\/Picture1.png?ssl=1\"><img data-recalc-dims=\"1\" decoding=\"async\" width=\"625\" height=\"455\" loading=\"lazy\" src=\"https:\/\/i0.wp.com\/www.blopig.com\/blog\/wp-content\/uploads\/2025\/09\/Picture1.png?resize=625%2C455&#038;ssl=1\" alt=\"\" class=\"wp-image-13094\" srcset=\"https:\/\/i0.wp.com\/www.blopig.com\/blog\/wp-content\/uploads\/2025\/09\/Picture1.png?w=860&amp;ssl=1 860w, https:\/\/i0.wp.com\/www.blopig.com\/blog\/wp-content\/uploads\/2025\/09\/Picture1.png?resize=300%2C218&amp;ssl=1 300w, https:\/\/i0.wp.com\/www.blopig.com\/blog\/wp-content\/uploads\/2025\/09\/Picture1.png?resize=768%2C559&amp;ssl=1 768w, https:\/\/i0.wp.com\/www.blopig.com\/blog\/wp-content\/uploads\/2025\/09\/Picture1.png?resize=624%2C454&amp;ssl=1 624w\" sizes=\"auto, (max-width: 625px) 100vw, 625px\" \/><\/a><figcaption class=\"wp-element-caption\">A flock of OPIGlets taking in the Fenway Park experience at a Red Sox game.<\/figcaption><\/figure>\n<\/div>\n\n\n<!--more-->\n\n\n\n<p class=\"wp-block-paragraph\">GRC CADD offered an incredibly valuable and intimate setting to explore the current state and future directions of the field. With a relatively small group of attendees and a strong emphasis on open discussion, the conference provided a rare opportunity to connect with both established experts and fellow early-stage researchers in an environment that encouraged deep, meaningful conversation.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">A wide breadth of topics was discussed, from quantum mechanics and ML potentials like ANI, to structure prediction and docking tools such as Boltz-2. This led to interesting discussions around generalisation and memorisation in such models.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">While much of the work presented was proprietary and, several exciting projects from the accompanying Gordon Research Seminar (GRS) were publicly available. To give a flavour of the diversity on show, some of the presented work included:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u2018SynLlama \u2013 Generating Synthesizable Molecules and Their Analogs with Large Language Models\u2019, which used large language models for synthesis-aware molecule design;<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u2018DeepPath \u2013 Overcoming data scarcity for protein transition pathway prediction using physics-based deep learning, which explored protein dynamics in data-sparse regimes\u2019, combining physics and ML to generate transition pathways between conformational states; and<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u2018An algorithmic framework for synthetic cost-aware decision making in molecular design\u2019 (SPARROW), which proposed a principled framework for molecule design under a range of practical constraints, such as synthetic cost.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">As a first-year PhD student, the conference was a brilliant environment to meet leaders in the field, learn from them, and exchange ideas. In fact, the coffee breaks and informal conversations often proved just as insightful as the formal talks themselves.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>Fergus<\/strong><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">While the latest \u201cco-folding\u201d and structure prediction methods (and several very cool <a href=\"https:\/\/www.nature.com\/articles\/s41467-024-51771-2\">extensions<\/a> and <a href=\"https:\/\/www.nature.com\/articles\/s41467-025-60759-5\">applications<\/a>) were at the center of many discussions throughout the week, the show was well and truly stolen by a single figure.<\/p>\n\n\n<div class=\"wp-block-image is-style-default\">\n<figure class=\"aligncenter size-full is-resized\"><a href=\"https:\/\/i0.wp.com\/www.blopig.com\/blog\/wp-content\/uploads\/2025\/09\/Picture2.png?ssl=1\"><img data-recalc-dims=\"1\" decoding=\"async\" width=\"625\" height=\"582\" loading=\"lazy\" src=\"https:\/\/i0.wp.com\/www.blopig.com\/blog\/wp-content\/uploads\/2025\/09\/Picture2.png?resize=625%2C582&#038;ssl=1\" alt=\"\" class=\"wp-image-13095\" style=\"width:400px\" srcset=\"https:\/\/i0.wp.com\/www.blopig.com\/blog\/wp-content\/uploads\/2025\/09\/Picture2.png?w=778&amp;ssl=1 778w, https:\/\/i0.wp.com\/www.blopig.com\/blog\/wp-content\/uploads\/2025\/09\/Picture2.png?resize=300%2C279&amp;ssl=1 300w, https:\/\/i0.wp.com\/www.blopig.com\/blog\/wp-content\/uploads\/2025\/09\/Picture2.png?resize=768%2C715&amp;ssl=1 768w, https:\/\/i0.wp.com\/www.blopig.com\/blog\/wp-content\/uploads\/2025\/09\/Picture2.png?resize=624%2C581&amp;ssl=1 624w\" sizes=\"auto, (max-width: 625px) 100vw, 625px\" \/><\/a><figcaption class=\"wp-element-caption\">From the paper \u201cHave protein-ligand cofolding methods moved beyond memorisation?\u201d (aka Runs\u2019n\u2019Poses), this is part of \u201cFig. 1 Prediction accuracy vs. training set similarity\u201d and shows that co-folding methods do quite a lot better when making predictions on complexes that are similar (similarity defined as the product of the binding pocket coverage and Combined Overlap Score of the ligand pose).<\/figcaption><\/figure>\n<\/div>\n\n\n<p class=\"wp-block-paragraph\">These results are a stark reminder to the field of the limits of computational techniques, especially those that learn from data, and clearly demonstrate the reliance of methods on the training data. Conversations largely focused on which region(s) of the graph were most important in drug discovery and how we could improve things going forward. While I think this is incredibly powerful, useful work, the headline result probably should not come as a surprise. Further, on some level, that we can correctly predict over 20% of complexes (in the case of AF3) with less than 20% similarity to anything we\u2019ve ever seen before is fairly impressive (I have not seen a direct comparison to non-cofolding methods). It does, however, show have far we have left to come, and I\u2019m excited for the many updates that will inevitable follow in the coming years (and be presented at the next CADD GRC!).<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>Matteo<\/strong><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Several talks address the issue of how we are training and validating our models. This includes the necessity to include more experimental data in the benchmarking dataset, as well as proper strategies for creating validation sets (especially for docking methods). This topic is becoming increasingly important as we move into an era dominated by less explainable architectures that can easily memorise or exploit flaws in our training and validation sets.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Two main problems have been highlighted that need to be addressed: (i) standardising the way we test new methods, given that we currently test the accuracy of new work on limited systems and with ad hoc datasets, often forgetting to compare the accuracy of the model with simple baselines or other methods This also included the use of proper statistics on the reported results, rather than tables with single numbers, as confidence intervals can give us an idea of how the methods are really improving compared to others.; and (ii) removing possible flaws from our current validation dataset. This is important because, even though training and validation sets are usually claimed to be completely independent and not overlapping, this is often not the case in practice, since the constraints used to define the two datasets are often too weak. An example of this is the use of the PDB with a temporal split for training and validation, since new structures in the PDB usually have homologues that were resolved years earlier and are present in the training set. Therefore, we should focus on including more filters to separate the validation and training sets. One possible way is to use separation methods that employ strict structural similarity filters and take homologous sequences into account.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Martin<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">My first highlight is the talk of Olexander Isayev on the development of machine learning-based foundation neural network potentials. Building transferrable models is the bread and butter of force field engineers and after ANI and AIMNet, there are now multiple flavours of AIMNet2 for various applications including reaction modelling, protein-ligand affinity prediction, and conformation generation.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The second highlight was Christofer Tautermann speaking on the potential benefits of quantum computing in drug design. The main take away was that while quantum effects do play a role in drug design in torsional barriers, conformation search reactivity, and binding (free) energies amongst others, there are currently no systems where quantum effects seem to be the bottleneck of drug discovery. Running DFT calculations in less than 1 second would be the real game changer.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Overall, we all highly recommend the conference to any budding computational drug development scientist; attendance was evenly split between representatives from industry and academia, talks were mostly on computational methods and programs, and there was ample time to meet people and socialize. <\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Henry Back in July, some very lucky OPIGlets ventured across the pond to discover life in Southern Maine (and Boston!). For someone visiting Boston for the first time, no trip would be complete without a Red Sox game\u2014a thoroughly enjoyable highlight (see Figure 1). While we were there, we also went to Gordon Research Conference [&hellip;]<\/p>\n","protected":false},"author":126,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"nf_dc_page":"","wikipediapreview_detectlinks":true,"_monsterinsights_skip_tracking":false,"_monsterinsights_sitenote_active":false,"_monsterinsights_sitenote_note":"","_monsterinsights_sitenote_category":0,"ngg_post_thumbnail":0,"_jetpack_memberships_contains_paid_content":false,"footnotes":""},"categories":[52,849],"tags":[879,878,877],"ppma_author":[790,535,779,487],"class_list":["post-13091","post","type-post","status-publish","format-standard","hentry","category-conferences","category-drug-discovery","tag-879","tag-cadd","tag-grc"],"jetpack_featured_media_url":"","jetpack_sharing_enabled":true,"authors":[{"term_id":790,"user_id":126,"is_guest":0,"slug":"jamesb","display_name":"James Broster","avatar_url":"https:\/\/secure.gravatar.com\/avatar\/df390c53770be6a0afc152da99e17025226f7300979c8b5e54021ddeb87971e4?s=96&d=mm&r=g","0":null,"1":"","2":"","3":"","4":"","5":"","6":"","7":"","8":""},{"term_id":535,"user_id":50,"is_guest":0,"slug":"fergus2","display_name":"Fergus Imrie","avatar_url":"https:\/\/secure.gravatar.com\/avatar\/19c18fa7f4d0a2aecc5f69760c6a9f2fc9b493dfe45b1fd333ccb447db9d6a90?s=96&d=mm&r=g","0":null,"1":"","2":"","3":"","4":"","5":"","6":"","7":"","8":""},{"term_id":779,"user_id":123,"is_guest":0,"slug":"matteo2","display_name":"Matteo Cagiada","avatar_url":"https:\/\/secure.gravatar.com\/avatar\/24f6f6ee42cb01af17d0e2c2d116bdcb62ee4597d10fb32d70868bfdc061b370?s=96&d=mm&r=g","0":null,"1":"","2":"","3":"","4":"","5":"","6":"","7":"","8":""},{"term_id":487,"user_id":92,"is_guest":0,"slug":"martin","display_name":"Martin Buttenschoen","avatar_url":"https:\/\/secure.gravatar.com\/avatar\/766a8e998df1df02635f3d2411a8526569f394d114b2fc9ebb896d84bb37484f?s=96&d=mm&r=g","0":null,"1":"","2":"","3":"","4":"","5":"","6":"","7":"","8":""}],"_links":{"self":[{"href":"https:\/\/www.blopig.com\/blog\/wp-json\/wp\/v2\/posts\/13091","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.blopig.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.blopig.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.blopig.com\/blog\/wp-json\/wp\/v2\/users\/126"}],"replies":[{"embeddable":true,"href":"https:\/\/www.blopig.com\/blog\/wp-json\/wp\/v2\/comments?post=13091"}],"version-history":[{"count":5,"href":"https:\/\/www.blopig.com\/blog\/wp-json\/wp\/v2\/posts\/13091\/revisions"}],"predecessor-version":[{"id":13100,"href":"https:\/\/www.blopig.com\/blog\/wp-json\/wp\/v2\/posts\/13091\/revisions\/13100"}],"wp:attachment":[{"href":"https:\/\/www.blopig.com\/blog\/wp-json\/wp\/v2\/media?parent=13091"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.blopig.com\/blog\/wp-json\/wp\/v2\/categories?post=13091"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.blopig.com\/blog\/wp-json\/wp\/v2\/tags?post=13091"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/www.blopig.com\/blog\/wp-json\/wp\/v2\/ppma_author?post=13091"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}