Everywhere we look these days, we see AI. In computational and medicinal chemistry, it feels like every conversation eventually turns to what AI might do for drug discovery.

Of course, generating compound ideas is only part of the design process. AI can produce plenty of suggestions, and just like ideas from people, some are inspired and some leave you wondering what it was thinking.

The challenge lies in determining which ideas deserve attention. Do they offer the balance of properties we need? Is the chemistry something we want to pursue? For me, the bigger question is how we manage a glut of ideas without overlooking something promising or spending all our time reviewing suggestions we would never take seriously.

This is where generative chemistry within StarDrop becomes particularly interesting.

In StarDrop 7.2, we released Inspyra™, which combines generative chemistry with an Inference Engine that learns from your interactions with suggested compounds. The premise appealed to me: your judgment helps guide what suggestions you see next.

In StarDrop 8.3, we’ve strengthened that approach. Inspyra now learns more from each piece of feedback you give, and its suggestions are easier to assess at a glance, so more of your time and effort goes on ideas that deserve a closer look.

Inspyra learns more from your feedback

Medicinal chemistry design involves balancing several objectives at once. A compound may have attractive potency but unsuitable solubility, or an appealing property profile with a structural feature you would prefer to avoid. Our preferences depend on how these considerations fit together.

The enhanced Inspyra Inference Engine learns your preferences across multiple properties simultaneously, considering their relationships as it selects ideas to suggest. This better reflects the balance you’re trying to strike, which is much closer to how you would judge a compound yourself. It also learns your preferences for chemical fragments, steering suggestions towards structural motifs you find interesting.

Each time you interact (positively or negatively) with a structure in Inspyra, the Inference Engine learns more about what you really want to see, and its suggestions become more relevant to your project.

We have also changed the colouring of suggestion borders. Previously, a binary system showed whether an idea increased or decreased a selected property. Now, a red–yellow–green scale indicates the predicted likelihood that you will respond positively by liking an idea or adding it to your data set, based on your previous interactions with suggestions.

The colours reflect everything that Inspyra’s Inference Engine has learned about your preferences, across multiple properties and structural motifs, to answer a more useful question: is this an idea I’m likely to want to take forward? It provides a quick and intuitive cue to help you prioritise your review, and one that better reflects the multi-parameter nature of compound discovery.

More room and more control when reviewing ideas

Learning from feedback works best when reviewing and responding to ideas feels natural.

In StarDrop 8.3, you can:

• Resize the Inspyra Panel to give structures and property plots more room, choose how many ideas appear, and adjust the refresh interval to suit your pace
• Use property plots to compare each suggested compound’s calculated properties with all the other ideas
• Apply a feedback action to all displayed ideas at once, making it quicker to respond to a batch of suggestions and giving the Inference Engine more to learn from in a single step

Smoother everyday workflows in StarDrop

StarDrop 8.3 isn’t just about Inspyra learning more from your feedback – we’re also learning. Much of the wider release has been shaped by the requests and experiences you share with our application scientists, support team, and account managers. Those conversations help us understand where changes would make the biggest difference to your everyday work.

StarDrop 8.3 now allows you to have multiple independent instances of StarDrop side by side. You can work in separate workspaces simultaneously, keeping two projects or analyses open without repeatedly switching between them, whether that’s comparing a new series with an established one or keeping a reference analysis to hand while you design. I’ve been using this myself for a couple of months now and for the things I need to do, it’s a game-changer.

There are new options for presenting results, too, making it easier to share the reasoning behind a design decision with your project team. Transformations and Matched Pairs tables can now be exported directly to PowerPoint, while histograms and pie charts can be labelled with counts and percentages. Histograms containing multiple series now show each series in its own bar, making the distributions easier to compare.

For Cerella™ users, the Suggested Measurements dialogue includes a new Exploration focus setting. Decreasing it from the default value of 1 may produce more diverse selections, giving you another way to guide the choice of measurements.

More dependable calculation services

The services behind StarDrop are an important part of the experience. We have re-engineered job handling in the Model and Metabolism servers to improve reliability. The Model server is also simpler to scale as calculation demand grows.

Server configuration is easier for Optibrium-hosted services as well. When you sign into StarDrop, connection details for all Optibrium-hosted services are retrieved automatically, meaning you can get to work right away.

Shaped by your feedback

As with previous releases, many of these improvements respond to the practical needs of scientists using StarDrop – including GUI improvements in Inspyra.

This release also brings refinements to R-group decomposition, Collaboration, and other workflows that users have told us about. Our shorter development cycles help us put these changes into your hands sooner.

If you would like to explore the new Inspyra features or discuss how StarDrop 8.3 could support your team’s workflows, contact us at support@optibrium.com.

The release notes provide the full list of enhancements and fixes.