そーせいグループ 研究

Sosei Heptares : Chris de Graaf氏によるタンパク質-リガンド構造予測 プレゼンテーション(全文・和訳)

AI4Proteins: Protein-Ligand Structure Prediction for GPCR Drug Design
発表者:Dr Chris De Graaf, Director, Heatd of Computational Chemistry Sosei Heptares
会議:AI4Proteins: Protein Structure Prediction



Thank you for your patience. Thanks so happy to kick off after this standard disclaimer that I always have to show my company. So, the company SoseiHeptares has an R&D Center in Cambridge and headquarters in Tokyo and you should have discovered the difference and the different stages in Discovery with R&D Center really focusing on the early discovery and early development and late stage developments in Tokyo.




As Chris already mentioned, we have a StaR technology that aims to identify combinations of point mutations in GPCR to thermo-stabilize them to make them amiable(x) amenable for structure-based drug design to really allow screening by a physics experiment and also structure determination and to use a structure interactive manner in a design process. And this has generated so far more than 70 different StaRs in different GPCR and different ligand bound states.




And, it's truly a very exciting era to work on GPCR structure-based drug design. A lot of structures are being solved and also available in PDB. Many more also that we still have in-house some of those are being published once in a while. but it's a very nice coverage actually now and in the past. They say 10 years has really accelerated and really provide an view all these different small molecules and peptide and protein-link bind with this wonderful receptors. We really want to use the high-resolution of structures to optimize ligands which is really atom by atom optimization to identify possibilities to design polar context for example to improve properties but also the consideration of multiple structures is very important to anticipate the flexibility and selectivity a determinant. And, very important aspects of all these different structures and having axis to even more is that there are also a lot of very humbling experiences for computational chemists in prediction especially also the small molecule ligand binding modes. But we initially maybe thought that they are able to bind in sort of classical binding sites accessible from the extracellular range. And we now realize actually ligands bind all over the places that is a very important challenge in our efforts.

また、GPCRの構造に基づくドラッグデザインに取り組むには、非常にエキサイティングな時代になっています。多くの構造が解明され、PDBにも掲載されています。また、社内にはまだ多くの構造があり、そのうちのいくつかはたまに発表されていますが、現在も過去も非常に素晴らしいカバー率です。10年前に比べて本当に加速しており、様々な小分子やペプチド、タンパク質リンクがこの素晴らしい受容体に結合する様子を見ることができます。私たちは、構造の高解像度を利用してリガンドを最適化したいと考えています。それは、例えば特性を向上させるために極性のある構造を設計する可能性を特定するための、まさに原子ごとの最適化ですが、柔軟性や選択性を決定づける要因を予測するためには、複数の構造を考慮することも非常に重要です。そして、これらの異なる構造の非常に重要な側面と、さらに多くの軸を持つことは、計算化学者にとって、特に低分子リガンドの結合様式を予測する上で、非常に謙虚な(x) 謙遜してしまうような経験がたくさんあるということです。私たちは当初、リガンドは細胞外からアクセスできる古典的な結合部位に結合できると考えていました。しかし今では、実際にはリガンドはあらゆる場所に結合することがわかっており、これは我々の努力の中で非常に重要な課題となっています



So, I would like to focus first on this aspect. We really focus the structure prediction elements really on the structure GPCR ligands interaction parts and then go into the importance of devils considering details in structure-based drug design and then at the end actually how can we combine all these this information from the structure point of view and from a pharmacological point of view to navigate this interested(x) interest across the GPCRome.

そこで、まずこの点に焦点を当ててみたいと思います。私たちは構造予測の要素をGPCRリガンドの相互作用の部分に焦点を当て、構造ベースのドラッグデザインにおける細部を考慮する悪魔の重要性に踏み込み、最後に実際に、構造的観点と薬理学的観点からのこれらの情報をすべて組み合わせて、GPCRome全体に興味を持ってもらえるようにナビゲートする方法を考えます。(x) この興味をGPCRomeの中へナビゲートします。



Yes. So, maybe a little bit back to a competition in the GPCR field on predicting GPCR structures and GPCR ligand interaction was really at the era not that many structures available. And, crystallographer CXCR4 challenge all the GPCR models around the world to solve the structure bound to molecule ligands. Yeah, we had a sequence. We had few templates. We knew what the small molecule ligands look like. But, very importantly there was a lot of experimental data like for example they're simply summarized here mutation data and when you have a look at this mutation data that actually you can get to, you know, a sort of this simple hypothesis where you can imagine that positively charged groups in a molecule can interact negatively charged residues around the binding sites. But we were faced by a so-called so what I could say Mr. Bean dilemma. So, this is the episode that he goes to the dentist. He knocks out the dentist so he has to take care of it himself. And, he sees his X-lay and he starts drilling but then he realizes it actually can flip around the X-ray so then he starts drilling on the other side of his mouth. Then actually he finds out that there's another way to turn around X-ray. Finally he ends up trying a lot of different tooth drilling at the dentist. We were facing a similar dilemma here because we have sort of two four symmetry in the binding site and also symmetry in the ligand so the different way that its molecule can fit.

さて、少し話は戻りますが、GPCRの分野では、GPCRの構造やGPCRのリガンド相互作用を予測する競争が行われていましたが、当時はまだそれほど多くの構造が出回っていませんでした。 そして、CXCR4の結晶学者が世界中のGPCRモデルで、分子リガンドと結合した構造を解くように挑戦しました。 そう、私たちには配列があったのです。テンプレートはほとんどありませんでした。低分子のリガンドがどんなものかはわかっていました。しかし、非常に重要なのは、多くの実験データがあったことです。例えば、ここには簡単にまとめられた突然変異データがあり、この突然変異データを見てみると、分子内の正電荷の基が結合部位周辺の負電荷の残基と相互作用すると想像できるような、ある種の単純な仮説を得ることができます。しかし、私たちはいわゆる、Mr.ビーンのジレンマに直面しました。ミスタービーンが歯医者に行ったときのエピソードです。彼は歯医者をノックアウトしてしまったので、自分で治療しなければならなくなりました。そして、X線写真を見て、ドリルで穴を開け始めるのですが、X線写真を裏返すことができることに気づき、反対側の口に穴を開け始めるのです。そして、X線を反転させる別の方法があることを知ったのです。最終的には、歯医者さんでいろいろな歯の穴あけを試してみることになります。今回も同じようなジレンマに直面していました。というのも、結合部位には2つの4つの対称性があり、リガンドにも対称性があるので、分子の収まり方が違うのです。



So that's what we did so we have the different experimental data. And, we couldn’t be very conclusive. So we propose different solutions and also look at SAR to try to at least optimize and prioritize if it would not be binding modes.




And then, in the end, we were actually quite happy that at least one of our solutions was correct. But, we were surprised at we were actually the only who have predicted this binding mode, which at least key interactions with right residues was correctly predicted.




AAnd end up actually putting it in a broader perspective on these GPCR dock competitions, what we actually found out it that prediction the folds and overall team helical confirmation. Perhaps it's not such a big challenge especially now with more and more templates being available that harmonicas(x) homologous to GPCR which now structures we solved. What is really challenging is the prediction of the contacts(x) contexts, the prediction of the binding modes of the small molecule ligands which can be very challenging.

結局、GPCRのドッキングコンテストについて、より広い視野で考えてみると、実際に分かったのはフォールドの予測とらせん全体の確認でした。 GPCRと調和した構造(x) 相同な構造を持つテンプレートがどんどん出てきている現在では、それはそれほど大きな課題ではないかもしれません。本当に難しいのは、接触(x) コンテクスト(前後関係)の予測や、低分子リガンドの結合モードの予測で、これは非常に難しいことです。



So, just to summarize the protein structure prediction. The helical folds are feasible. The loops are difficult to predict despite conserving a dozen of bridges. But, over for despite feasible, what is a very challenging is first of all predicting the binding sites. And, there are a lot of surprises and I continue to surprise in this area but also GPCR ligand binding mode predictions. And, I think Thomas mentioned earlier today. Binding site solvation is important, conformational changes, but also GPCR the role of the membrane lipids which is another standard think to consider at least in the initial construction of these receptors and complexes.

ということで、タンパク質の構造予測について、ちょっとまとめてみました。ヘリカルフォールド(らせんの折り畳み計上)は実現可能。ループは十数個の橋を保存しているにもかかわらず、予測するのは困難です。しかし、実現可能であるにもかかわらず、非常に困難なのは、まず結合部位の予測です。この分野だけでなく、GPCRリガンドの結合様式の予測についても、多くの驚きがあり、私自身も驚き続けています。今日、トーマスが言っていたと思いますが 結合部位の溶媒和は重要ですし、コンフォメーション(配置)の変化も重要です。また、GPCRの膜脂質の役割も、少なくともこれらの受容体や複合体を最初に構築する際に考慮すべき標準的な考え方です。



Just an example of the case actually where to fold is very difficult to predict. An orphan GPCR where also supposed to binding mode and conformation were actually very challenging to predict.



So now, dig into the details of the structure-based drug design. As so fully focusing on this part small changes atom by atom optimization steps and how we can use structures and computational approaches to achieve this.



A very important aspect here will be developing recipes together among the “Molecular Discovery” to analyze the binding sites and to identify lipophilic hotspots in the binding sites and drive binding and that also by displacing and rejecting unfavorable water molecules. And this is really an under-forgotten and under-represented aspect of the pharmacophore models and now looking at the multiple structures and looking at the ways the ligands actually align in those binding sites. We get a better view on the actual driving effectors of a GPCR ligand binding.
ここでの非常に重要な点は、結合部位を分析し、結合部位にある親油性のホットスポットを特定して結合を促進するためのレシピを、「Molecular Discovery」のメンバーと一緒に開発することです。これは、ファーマコフォアモデルの中でも特に忘れ去られていた側面であり、今では複数の構造を見て、リガンドが実際に結合部位に配置される方法を調べています。これにより、GPCRリガンド結合の実際の駆動エフェクターについて、より良い見解が得られます。



Water molecules are very important. They determine a binding affinity, selectivity and kinetics. And there are several computational methods available now including “WaterFLAP” that using also water map. That aim on assessing the energetics of these networks and also to assess the way that ligands can modulate or influence these networks.



It’s really the protein binding sites far more complex than a shape illustrated here by the overlay of the different ligands in these receptors. And, it’s really these lipophilic hotspots that seem to drive the binding of all these different types of ligands across different types of GPCR.



One way to assess and to look at the APRIL protein(x) apoprotein is to look at what are the regions in the binding sites that can be targeted in the design of optimization of drugs to displace on a happy water. Another way to assess this to look at the way that the ligands influence the water network and some cases stabilize unhappy water molecules. And, 3rd ways to look at the compared of way at close relatives sectors like here adenosine a2a, one(x) a1 receptor very small changes and a shape, and properties of binding sites. can I have a dramatic effects on the water networks. In this case, a1 trapping an unhappy water molecule in this pocket which is not present in the a2a receptor explain the selectivity of this molecule for the a2a receptors.
APRILタンパク質(x) アポタンパク質を評価する一つの方法は、結合部位の中で、幸せな水に置換する薬剤の最適化設計でターゲットとなる領域はどこかを調べることです。もう一つの評価方法は、リガンドが水のネットワークに影響を与え、場合によっては不幸な水分子を安定化させる方法を見ることです。そして、3つ目の方法は、このアデノシンa2aのような近縁種のセクターでの方法の比較を見ることです。1つの受容体の非常に小さな変化と形状、結合部位の特性は、水のネットワークに劇的な影響を与えることができます。この場合、a1は、a2a受容体には存在しない不幸な水分子をこのポケットに捕捉することで、この分子のa2a受容体に対する選択性を説明することができます。



Just to show the application of this view on a structure based drug design are going from an initial template where a long molecule is required and large molecules are required to the target of this site of the binding site. An opportunity to target another pockets in the a2a receptor identify the virtual screening and then after solving the crystal structure, and appreciate in the details of interactions to displace this unhappy water to really optimize and the features of this molecule and this is now the front running compound that is now collaboration to AstraZeneca has been developed now in immuno-oncology.



Another example of the other importance of lipophilic hotspots and water networks is nicely illustrated by multiple olexin receptor crystal structures in PDB and also source in-house and published last year. If you look at the alignmental ligands and waters and any interactions, you actually realized that no polar interaction of hydrogen bond donar acceptor pharmacophore, not a lot of the conserves polar interactions directly with the residues. But, there are water mediate interactions are very important for binding modes prediction of these ligands in the binding sites and also subtle changes in the water network also in this case that can explain selectivity of a , in this case, EMPA molecules for OX2, first OX1 where unhappy water molecules is trapped in the binding sites.
親油性のホットスポットと水のネットワークの重要性を示すもう一つの例が、PDBや社内資料で昨年発表された複数のオレクシン受容体の結晶構造にうまく示されています。配置されたリガンドと水、そして相互作用を見てみると、水素結合ドナー・アクセプター・ファーマコフォアのような極性相互作用はなく、残基と直接極性相互作用を保存しているものはあまりないことがわかります。 しかし、水を介した相互作用は、リガンドの結合部位での結合様式を予測する上で非常に重要であり、また、水のネットワークの微妙な変化も、今回のケースでは、不幸な水分子が結合部位に閉じ込められているOX1から、OX2に対するEMPA分子の選択性を説明することができます。



Taking the interesting devil’s in the details, one step further is the use of free energy perturbation approaches where we have been corroborating with which learning(x) Schrödinger to get a FEP approach to optimize that for GPCR structure-based drug design. A very important aspect here that the method should not be used directly out of the box without thinking automated manner. It’s really about the careful setup of the system, optimization of the system, step by step considering binding site solvation, membrane also, and sufficient sampling of the binding sites.
GPCRの構造に基づいた薬剤設計に最適なFEPアプローチを得るために、どのような学習をしたかを確認してきましたが、興味深い「悪魔の細部」をさらに一歩進めて、自由エネルギー摂動アプローチを使用しています。(x) 自由エネルギー摂動法(FEP法:結合シミュレーションのための計算手法の一種)を用いて、シュレーディンガーと協力してFEP法を確立し、GPCRの構造に基づく薬物設計に最適化しています。ここで非常に重要なことは、この手法を自動化を考えずにいきなり使うべきではないということです。慎重にシステムをセットアップし、システムを最適化し、結合部位の溶媒和、膜、結合部位の十分なサンプリングを段階的に考慮することが重要なのです。



Here’s just initial steps showing how the GCMC setup is really important to identify at the start. The right ligand dependents binding site solvation to avoid, to optimize the model and get up correct ranking of meso-philic(x) experimental affinities for related ligands. And another aspects is also careful consideration of protonation state of residues in the binding sites are like as he study(x) histidine that the really influence is also that structure and complex as well as the standard ranking of different series in the FEP predictions.

ここでは、GCMC(グランドカノニカルモンテカルロ シミュレーション法)のセットアップがいかに最初に特定することが重要であるかを示す、最初のステップをご紹介します。モデルを最適化し、関連するリガンドの親和性の正しいランキングを得るために、避けるべき正しいリガンド依存の結合部位の溶媒和を示しています。また、もう一つの側面として、ヒスチジンのような(○)結合部位の残基のプロトン(陽電子)化状態を慎重に考慮することがあります。彼の研究によると(x)、FEP予測における異なる系列の標準的なランキングと同様に、構造と複合体にも影響を与えます。



And, an overview of another application in-house anomalies in this case, another targets were actually the consideration of different tautomer states of the ligand was also really important. And, I have to be incorporated in the FEP, a thermal demographic cycle for extra-prediction as well as a ligand depends conformational changes.



So, another changing gear for the last theme of the presentation. So how can we now combine this wonderful new insight in GPCR ligand structure with pharmacological data and ligand information to extrapolate to this information to other receptors.



And, how can we use especially also all these very surprising binding modes which have been very difficult to predict to other systems.



A very simple way to do that is to look at ligand similarity and to use new insight into binding sites to classify the putative binding sites of ligands for a certain receptor or to combine with sequence analysis where sequence analysis in this case can use to highlights positions of residues in specific receptors that have a high entropy outside systemly, and low entropy insides a sort of a family highlight in potential binding sites.



For a experimental point of view, also the mapping of mutational effects in pharmacological experiments on the binding sites in combination with the binding mode prediction can be a very powerful approach to get to higher resolution GPCR ligands models and in this case StaR generation program often offer this binding site mutation maps can use in-house.



Description of the different binding sites using the GRID approach that dimension earlier is also way to compare binding sites with those that predict protein ligands interaction and look for to several presentation actually also in this area. And, we actually had initiated a collaboration with them ”Molecular Discovery” to look at GRID-base to the answer in this area.
先に述べたGRID(Molecular Discovery社のソフトウェア)アプローチを用いて異なる結合部位を記述することは、結合部位をタンパク質リガンド相互作用を予測するものと比較する方法でもあり、実際にこの分野でもいくつかの発表が行われています。そして、私たちは実際に彼ら「Molecular Discovery」とコラボレーションを開始し、この分野での答えをGRIDベースで探していました。



Another message to actually reflects molection and experimental information to support GPCR ligand modeling approaches also to use this information for virtuous ligand screening so-called protein ligand interaction fingerprints which drives a simple interaction pattern a bit string from protein complex in the space from a crystal structure for example which can compare to the interaction bit strings of binding poses predict for other molecules to identify complete(x) chemically different molecules that can dock similar to binding mode.
もう一つのメッセージは、GPCRリガンドのモデリングアプローチをサポートするために、実際に選択と実験の情報を反映させ、この情報を好ましいリガンドのスクリーニングに使用することです。いわゆるタンパク質リガンド相互作用フィンガープリントと呼ばれるもので、これは、例えば結晶構造から得られる空間内のタンパク質複合体からの単純な相互作用パターンのビットストリングを駆動し、他の分子に対して予測される結合ポーズの相互作用ビットストリングと比較して、結合モードに類似したドッキングが可能な完全に(x) 化学的に異なる分子を特定することができます。



And, looking at this approach have been successfully applied in several GPCRs achieving high virtuous screening hit rate in terms of experimentally validated the hits for several receptors as to described here. What is very important also is that this is really direct way using a binding mode information experimentally or binding mode hypothesis to that allow someone, if validated, also allows a quick optimization, efficient optimization of the ligand-based on the hypothesis.



So, lots of pieces of the work initiate collaboration with the University of Cambridge to combination of the new generative machine learning. AI approach is to generate molecules Denovo combining that with different computer drug design approaches of ligand-based and structure-based approaches. And, interestingly, this approach is also in the class of virtual screening, highly complementary in sampling different parts of chemical space and interestingly, also in this case, generating molecules that can be consistently docks in line with binding modes of zero for(x) ,therefore, non-active molecules.



So, that’s all what I want to share with you and I’m very happy to follow the discussion and perhaps sharing also experiences in the application of new messages in a prediction protein ligands interactions



-そーせいグループ, 研究

Copyright© Shiroshiroのバイオ株ノート , 2023 All Rights Reserved Powered by AFFINGER5.