2022 AMI Online Salon

Predicting Structures

Project Details

  • Entrant Name:  Valerie Altounian
  • Client: AAAS/Science
  • Copyright: AAAS, 2021
  • Medium/software used: Cinema 4D, Photoshop
  • Final presentation format: printed journal cover and online issue
  • Primary Audience: Readers of Science, protein folding community

Project Description

This illustration pairs with a paper reporting a deep learning solution to the protein folding problem. Similar to the DeepMind framework, Baek et al’s algorithm, RoseTTA fold, can output accurate models of proteins. Novel to their work is the consideration of three tracks of information: sequence (1D), distance (2D), and coordinate (3D) information. Here, the folded structure of one of the authors’ test proteins is magically revealed out of a streamer representing these tracks. 1D data are illustrated by letters of this protein’s real amino acid sequence. Real contact map iterations from the protein make up the 2nd track, representing how close every amino acid is from each other. A dotted line that pops in and out of the third track abstractly illustrates the 3D placement of molecules.This illustration needed to reflect the magnitude and wonder of solving the protein folding problem. The composition and shimmering effects work to that end. The color palette was chosen to give the feel of being inside a computer network. This is also helped by the dark background with subtle gridlines reminiscent of 3D computer graphics. The lighting strikes a balance between a technological feel and real-world light. Offset colors along the edges of the protein as well as the movement of the tracks up and around it give the effect of the protein actively brought into existence by computing power.