2023-10-27 Flatiron Institute / Michael Shirts Meeting notes

Scientific / Technical Presentation

Vikram Mulligan

• Previous work: Designing peptides of mixed L/D amino acids with unnatural folds

• Macrocycles

  • Macrocycles designed to target the Wuhan variant of SARS-CoV-2 spike glycoprotein

    • micromolar affinity

    • Redesigned to target Omicron

• Current software limitations

  • Accuracy

  • Generality and extensibility

  • Tractability and efficiency

• Rosetta is a kinematics engine and a scoring engine

  • That is very good at building protocols

• Rosetta is very good at solving the rotamer optimization problem

  • This is how we do design

• Predicted fold propensity (estimated dG of folding) correlated well with IC50

• Mike Gilson: Mining Minima

• Enhancing Rosetta’s energy calculations with quantum chemistry

• Limitations of Rosetta as a platform for future science

  • MPI-parallel but not GPU-parallel

  • “Pose” data structure is optimized for side chain optimization, relaxation, and backbone sampling

    • Ideally we could convert data structures for different tasks

  • Rosetta is a monolith (software engineering perspective)

• Designing and building the Masala software suite

  • Usable as stand-alone or robust library

  • Free and open source successor to Rosetta under development at Flatiron

    • aGPLv3

  • Structured to take advantage of massively parallel CPU or GPU

    • Thread manager

    • Scales nicely up to hundreds of CPU cores

  • Versatile plugin architecture

    • Internal library

    • API layer

    • Plugins

    • User interfaces

  • API layer is built from JSON API description

    • Allows regression testing of API description

  • Special types of Masala plugins

    • Engines

    • Data Representations

    • These are usually handled in one plugin

      • define a new data representation optimized for a given engine

      • Convert data structure

      • pass it to engine

      • do the (highly efficient) work

      • Convert back to standard data structure

      • pass back to Masala internals

  • Using Masala to implement packing on quantum computers

• Hope to implement translator to and from OpenFF to set up force fields

  • Option 1: Implement internal Masala representation of molecule and implement our own machinery to parameterize it with force field from offxml

  • Option 2: Use OpenFF’s implementation (toolkit) to parameterize the system

  • Option 3: Use Interchange to output in a different format