1 of 15

Stability analysis of conformational landscapes with FlexConsensus

David Herreros, PhD Student, Biocomputing Unit (CNB)

2 of 15

How good are our measurements?

  • We are estimating our conformational landscapes directly from the experimental CryoEM data (particles, maps, structural models…)
  • The accuracy of the estimated landscapes relies on how well we can identify the structural features of a given state

How far are we from recovering our teapot landscape from experimental data?

Ideal landscape

3 of 15

How good are our measurements?

Ideal landscape

Estimated landscape

The landscape we are estimating is far from being the ideal one... Can we solve it?

4 of 15

How good are our measurements?

 

 

 

Bias

(systematic)

Covariance (random)

5 of 15

How good are our measurements?

 

 

 

Bias

(systematic)

Covariance (random)

6 of 15

How good are our measurements?

 

 

 

Bias

(systematic)

Covariance (random)

7 of 15

How good are our measurements?

 

 

 

Bias

(systematic)

Covariance (random)

?

8 of 15

Dealing with random errors

Estimated landscapes (independent runs)

Consensus landscapes

9 of 15

Dealing with random errors

Consensus landscapes

  • We need to correct first for systematic errors (CTF corruption, centering…)
  • Random errors cannot be corrected, but they are centered around the ideal solution
  • Thus, we cannot trust a single estimation, we need to do many of them and perform a consensus step to get closer to the ideal solution

10 of 15

How does FlexConsensus work?

Encoder Zernike3D

Loss

Decoder Zernike3D

Zernike3D bottleneck

Consensus space

Encoder CryoDRGN

Decoder CryoDRGN

Zernike3D space

CryoDRGN space

Predicted Zernike3D space

Predicted CryoDRGN space

CryoDRGN bottleneck

Encoder loss

Interactive stability analysis

11 of 15

How does FlexConsensus work?

CryoDRGN (R2)

CryoDRGN (R1)

Particles

FlexConsensus

Zernike3D (R2)

Zernike3D (R1)

12 of 15

How does FlexConsensus work?

Zernike3D (R2)

Zernike3D (R1)

Particles

FlexConsensus

CryoDRGN (R2)

CryoDRGN (R1)

13 of 15

How does FlexConsensus work?

Particles

Zernike3D (R2)

Zernike3D (R1)

CryoDRGN (R2)

CryoDRGN (R1)

FlexConsensus

14 of 15

How does FlexConsensus work?

FlexConsensus

15 of 15

Any questions?