_index.org

RoseTTAFold2

Last edited: August 8, 2025

RoseTTAFold2 is a three-track folding tool, which also handles multimer!

  1. inputs: amino acid sequence + CHEMICAL structure (WOAH! how?)
  2. “RF2 all-atom embedding”
  3. fold!

The model does really well!

application: de-novo luciferase design

  1. come up with the correct shaped scaffolds
  2. use old Rosetta to jam a residue sequence into the scaffold
  3. refold

application: RoseTTAFold2 in-painting

Train the model to recover the missing bits of sequence from the overall structure (i.e. training backwards), and

Rossing 1990

Last edited: August 8, 2025

On the dynamics of Tuning Forks. (Rossing, Russell, and Brown 1992)

Characterizing Tuning Forks

Aluminum, tines 10mm apart. Four main groups of vibration:

  1. Symmetrical In-Plane
  2. Antisymmetrical In-Plane
  3. Symmetrical Out-Of-Plane
  4. Antisymmetrical Out-Of-Plane

(a) and (c) are in the first group; (b) is in the second group, where the fork just warps.

Deriving Tuning Forks’ Frequency

As per before, we can treat tuning forks acting in clang and fundamental modes as a good’ol fashioned cantilever beam.

rotational energy theorem

Last edited: August 8, 2025

total kinetic energy

\begin{equation} KE_{rigid} = \frac{1}{2} M{V_{cm}}^2 + \frac{1}{2} I_{CM}{\omega_{CM}}^2 \end{equation}

torque from gravity

For even non rigid bodies, the following follows:

\begin{equation} \vec{\tau}_g = \vec{R}_{CM} \times M\vec{g} \end{equation}

Actually, this follows for any \(f\) (like \(g\)) evenly applied across point masses.

potential energy

\begin{equation} \Delta PE_g = mg\Delta h \end{equation}

where, \(\Delta h\) is the travel of center of mass. Regardless of whether or not its point.

RP

Last edited: August 8, 2025

runtime monitoring

Last edited: August 8, 2025

Goal: flags situations that might be hazardous when they occur to trigger some fallback mechanism.

Three steps:

  1. check if you are even in the design: operational design domain monitoring
  2. even if you are, check how certain you are: uncertainty quantification
  3. once you know what your model is going to do: failure monitoring

operational design domain monitoring

“Out of distribution detection.” Insight: if we operate outside of the ODD, our validation results may no longer be valid, so we should monitor whether or not we are in the operational design monitor whether or not we are in the operational design domain.