Developing Chemistry Apps in the Cloud
Contributed Commentary by John D. Clark, Chris Leeding, and Ed Champness
February 20, 2015 | The rapid incorporation of touch devices, such as smart phones and tablets, into our personal and business lives increases pressure on software providers to deliver technology via mobile apps. In some situations this means developing a mobile version of an existing product that is designed specifically for a touch device. Sometimes the mobile offering is an equivalent product; other times it’s a scaled-back version of the full product that provides only the most important features, or those most suitable for a touch environment.
We were very keen to develop a new iPad app for interactive compound design in drug discovery, which would combine chemical drawing tools with predictive models. We faced a number of interesting challenges designing this functionality for a mobile environment.
First, user interfaces designed to enable experienced scientists to interact with their molecules often require the presence of a mouse, a keyboard, and a highly complex menu structure. Thus, we had to design our app—Asteris—to use simple gestures that could be conveyed through a touch screen.
The most challenging aspect of designing the Asteris user interface was to enable drawing and editing molecule structures quickly and easily. While a finger motion can approximate the movement of the mouse, and tapping can replace the left mouse button, those gestures are less accurate than a mouse, and there is no notion of “hover-over” for cues about on-screen objects. In desktop software, the keyboard is often used when editing molecules and while it is possible to display an on-screen keyboard, we felt that this method of input would have been cumbersome, occupying a large proportion of the available screen space.
Rather than requiring accurate touches to create atom locations on screen, we chose to interpret gestures in a chemical structure domain. Many of the common gestures are used: a pinch to change the scale, a tap to select an object and a drag to move an on-screen object. We also created a few custom gestures to enable the user to draw complex molecules easily. For rings, a common structural element in molecules, a circular gesture was the most obvious choice. A circle gesture with two fingers creates an aromatic ring. A swipe gesture adds chains of a specified length of atoms, but we extended the swipe gesture to six directions to create saw-tooth chains of atoms in standard directions.
With a tap to select atoms and bonds, and with circle and swipe gestures interpreted to add rings and chains, it was no longer necessary to have the point accuracy that is provided with a mouse, and the drawn structures are created with mathematical accuracy.
Another challenge we faced was the knowledge that many drug discovery organizations develop their own predictive models of ADME (absorption, distribution, metabolism, and elimination), physicochemical, and other biological properties. While we wanted to incorporate our own models of a number of these properties to ensure Asteris would provide a comprehensive design environment out of the box, we did not want its utility to depend solely upon our own models. We felt it was important to ensure that drug discovery organizations could use Asteris with their own predictive models as well.
Therefore, to provide this flexibility we decided to develop a model server with which Asteris could communicate to generate predictions for molecules. To provide a publicly-available service we would then deploy this as a custom cloud service. The prevalence of Wi-Fi and mobile data connections ensures that devices remain connected to the internet and a cloud-based solution is accessible for most users. An advantage of this approach is that the computational power available is easily scalable. For Asteris, we use Amazon Web Services (AWS) which provides many options for cloud-based solutions. With AWS we were able to set up a web service that scales the number of computational units in the background, automatically varying according to the demand. At the same time, we configured operational rules so that a minimal level of service is guaranteed without committing unnecessary resources.
A final consideration for apps used in pharmaceutical design is security of intellectual property. A successful molecule is potentially worth billions of dollars to the organization making the discovery, and therefore the design and optimization of a molecule is highly guarded. As such, an app which connects to the internet (to the cloud) must protect the information that it contains.
The challenge presented by using a cloud-based solution is that it requires sending potentially sensitive information about molecules over the internet. Secure connections using the Hyper-Text Transfer Protocol (HTTP) with a Secure Sockets Layer (SSL) on top (often referred to as HTTPS) are now commonly used to send personal and financial data safely. This protocol ensures that the information exchanged between client and server is encrypted so even if someone is “watching” the transfer it will be very difficult to decode.
The Asteris servers are set up similarly, using HTTPS for all communication. The molecular structure must be communicated to the computational servers to predict the physicochemical and ADME properties. The predictive models currently available through Asteris only depend upon the two dimensional molecule structure which can easily be conveyed using the Simplified Molecular Input Line Entry System (SMILES). SMILES enables a molecule’s structure to be represented by a single string of characters, which is easily encrypted. The Asteris servers can then make very fast predictions for all the physicochemical and ADME properties and return the results back to the user via the secure connection.
The result is a customizable, interactive iPad app in which the user can design molecules and calculate properties which can be used to guide further molecular designs.
John D. Clark Ph.D., CSO at Integrated Chemistry Design; Chris Leeding, Ph.D. Associate Director for Product Development at Optibrium; and Ed Champness, CSO at Optibrium.