The ‘Computer-Aided Drug Design’ Department provides computational design support for our clients’ drug discovery projects. In this interview, we introduce our CADD experts and see what exciting developments are awaiting us this year.
Eddy Damen did his PhD in Nijmegen, the Netherlands and joined the company as an organic chemist 24 years ago. After being involved in building the Parallel Chemistry Department in the early 2000s, Eddy changed departments in around 2012, when medicinal chemistry including computational chemistry services started to grow,. Eddy started CADD support by learning on the job.
In 2021, Giovanni Bolcato joined the team. He studied medicinal chemistry and pharmaceutical technologies in Padova, Italy and did his PhD in pharmaceutical sciences, working in the field of computational chemistry. He worked with a group in Sweden, on a project about the application and development of artificial intelligence (AI) methods in drug discovery before joining Symeres.
“CADD at Symeres is mainly about generating ideas and understanding which ideas will efficiently move the project in the right direction” explained Eddy. CADD can be divided into ligand- and protein structure-based methods . Protein structure-based design, where you design your compounds based on the protein or receptor you’re targeting, makes up the vast majority of daily work, since nowadays more and more 3D structural data, such as crystal structures and (AI-derived) homology models, are available. With this structural information, CADD can help medicinal chemists optimize the compounds, primarily for potency and selectivity, but also for secondary properties such as improving ADME. A close interaction between the medicinal chemists in the lab and the CADD group and its tools are vital.
At the moment, the CADD Department uses the software suites MOE, from the Chemical Computing Group, and Maestro, from Schrödinger. Furthermore, they have access to Spark (Cresset), which has been quite successful in several projects where scaffold hopping was requested. Finally, LiveDesign™ (Schrödinger) can be used to share project data and computational models.
As well as collaborating on a range of projects using different computational approaches, the CADD team offers virtual screening. In virtual screening, a range of chemical structures are tested on the computer model of the intended target to see which structures theoretically would bind to the target. Under the right conditions and with sufficient rigor, it can greatly reduce the number of compounds that need to be screened through an actual biochemical assay, saving the customer both time and money. For this, a complete workflow has been established, from selection and preparation of a database to screening and nomination of the final hits. At Symeres, virtual screening can be performed using either Databases of millions of on stock compounds from commercial suppliers, the Symegold library (75,000 compounds) available at Axxam for high-throughput screening, or the Symeres proprietary Virtual Library. At the moment, this last library contains about 17 million compounds derived from unique scaffolds, which have been validated for library synthesis. Also, a diverse subset of commercially available virtual libraries has been prepared and can be used off the shelf in virtual screening.
One definition of real creativity is transforming original design ideas into successful assets. The team believes that the virtual screening offered at Symeres is unique compared to the virtual screening at many other companies. “While most companies only return hits based on a particular scoring and ranking in silico, we at Symeres strongly believe that integrating a substantial amount of human intelligence, i.e., the significant experience of our medicinal and synthetic chemists adds significant value in transforming ideas into practice”
Another recent highlight is the introduction of free-energy perturbation (FEP+). FEP+ is a computational method based on thermodynamics to calculate and predict the relative binding free energy of a ligand in a binding site. These calculations are computationally very demanding and have to be performed on a cluster of GPUs. When applied in the right way, FEP+ can positively contribute to lead-optimization projects by prioritizing compounds for syntheses.
Although worldwide, significant investments are being made in AI and machine learning, Giovanni and Eddy predict that eventually only companies combining outstanding techniques with robust training data will be successful. “machine learning is dependent on ‘big data’ and performs best when the algorithms and models are fed with a large amount of high-quality data,” says Giovanni. “However, this kind of data is only sporadically available in the public domain and access to this level of quality will be critical”
Eddy adds, “This lack of data is especially true for compounds that enter clinical trials, and those compounds are the ones machine learning could learn the most from.” Nevertheless, machine learning will have its place in the toolbox and the next five years will be critical for its true value to be demonstrated.
We would like to thank Eddy and Giovanni for their contribution to this article. If you want to receive updates on their progress in 2023, follow us on LinkedIn or subscribe to our newsletter.