Vision Arcanum Blog
InkToMolecule Open Beta
This is serotonin - a “don’t panic yet” neurotransmitter. To be precise, this image is a structural formula of serotonin, drawn by hand on a tablet with a digital stylus.
Let’s say, for whatever reason, you want to know its molecular weight or any other property. Of course, you can go to a chemical database and type “serotonin” in its search field, but what if it was some other, much less recognizable structure? No problem, simply redraw the structure using ChemDraw, MarvinSketch or any other similar tool of your choice and get its machine-readable representation (SDF, SMILES, InChI, etc.). Now you can find out everything about this structure! It is still a pity though, that you cannot simply drag-and-drop a sketch of a structure into a search field... Or can you?
Today we are happy to make a small step towards more natural interaction between humans and computers in the context of Chemistry. Today we launch an open beta of our very first tool for recognition of chemical structures drawn by hand. We call it simply InkToMolecule, and it is an add-in (extension) for Microsoft OneNote for Windows. It allows you to instantly convert sketches of structures into clean images and SDF files.
Imagine you are taking a digital handwritten note of a chemistry lecture. Hopefully the image of lecture notes does not evoke any unpleasant memories, otherwise we are very sorry for bringing this up… With InkToMolecule you can quickly replace all messy sketches with neat, computer-generated images. Or let’s say your sketches are perfectly fine already, but you want to add some details to the notes. Simply recognize a sketch, save its SDF file and use it to find more information about the structure. No need to redraw anything or even leave OneNote.
InkToMolecule is powered by our in-house recognition technology based on Deep Learning. This means that the quality of recognition will improve over time as we collect feedback from the users. We also plan to teach our models new tricks, for instance, add support for markush structures and reactions. But as of today, both the add-in and the core technology are in early access.
We are the first to admit, that the current state of recognition accuracy leaves much to be desired. However, we can assure you that it does not reflect the final quality that we are striving to achieve. During the open beta, which will last approximately six month, we are going to collect usage data that will help us to improve the models.
Since some drawings may be collected by our recognition server, please, do not use this version of InkToMolecule for any sensitive information.
We encourage you to embark on the early access train and give InkToMolecule a shot. Simply sign up for the open beta and download the installer. Make sure to check out the user guide. What if you have never used OneNote before, but still want to try InkToMolecule? Chances are, OneNote is already installed on your Windows computer as part of Microsoft Office, otherwise you can always download it for free.
Any feedback is highly appreciated, but perhaps the most valuable to us are your reports of recognition mistakes. If you notice that a structure is not recognized correctly, press “Report a mistake” button to notify our R&D team with a gentle electric discharge.