You are here


3 posts / 0 new
Last post

Hello Everyone,

I have been trying to find a solution to RosettaDesign where it can reliably design proteins.

I have tried the following protocols:

FIXBB with resfile   - faield
FIXBB                - faield
FLXBB with blueprint - failed
FLXBB                - failed
Remodel              - failed

In each protocol I choose the lowest scoring decoy from 100 generated decoys. I run an abinitio simulation (25,000 decoys) but I do not get a funnel shaped plot.


I have the following questions:

1. Which is better, fixbb or flxbb? which one should I concentrate my efforts into learning and perfecting?

2. Is it possible to design any ideal protein? (80<size<150, globular compact structure, and mostly helices and sheets rather than floppy loops) or does the design work for some proteins and not others? I am trying to RosettaDesign PDB ID 4WYH

3. Am I supposed to add an additional step that I am not adding?

I am following the demos to the word, but I am not getting good results.

When I analyse my lowest scoring structure it seem to be perfect, no voids, low energy, the correct amino acids in the correct places etc... what am I getting wrong?

What is the correct way to design proteins using rosetta?

Post Situation: 
Sun, 2019-07-21 01:20

When designing, an easy design task would be one with a success rate (that is, the fraction of designs that successfully fold when run through Rosetta ab initio structure prediction) of perhaps 5%.  Hard design tasks can have success rates of 1 in 1000.  Do not just forward fold the lowest-energy design.  Forward fold the top hundred to find a hit.

The reason for this is that the design process can only consider the designed conformation, and it tries to find a sequence that minimizes the energy of that conformation.  However, minimizing the energy of a desired state is not the same as maximizing the energy gap between that state and all alternative states: it is entirely possible that the design process stabilized an alternative state even more than it stabilized the designed state, for example.  Only a small fraction of the time does it happen to be the case that you uniquely stabilized the design state.  So a lot of guessing-and-checking is necessary.

As a final suggestion, look at design-centric guidance terms ( as a means of imposing prior information that is likely to raise the design success rate.  Naive Rosetta design makes foolish mistakes, like creating cores with too many alanines or too many bulky aromatic residues.  Terms like aa_compsition ( give you some control over the path that the design process takes, though, letting you coax it in a more sensible direction with directives like, "make sure that the core is 80% valine, leucine, and isoleucine" or whatnot.

Mon, 2019-07-22 13:03

Hi vmulligan,

Thank you for your explanation, that answered a lot of my questions and changed my expectations of the design outcome, and the links were very useful, thank you.

I have changed much of my setup (and have been running lots of benchmarks), but I do have 2 questions:

1. These are the weights I chose:

aa_composition                       1.00
netcharge                            1.00
aa_repeat                            1.00
aspartimide_penalty                  1.00
hbnet                                1.00    # gives bad fragment (test further later)
voids_penalty                        0.10    # gives bad fragment (test further later)
approximate_buried_unsat_penalty     5.00
buried_unsatisfied_penalty           0.75

I chose these wight values from reading about them, either the recomended value for each, or the midpoint value of the recomended range.
Do these values seems logical? is there a "best practice" weight values that are a good starting point?

2. Where can I find a more in depth explanation of the aa_composition? at the moment I cannot judge what % of which amino acid I need and how to determine what is good and what is bad. Where can I learn the science behind this concept? so I can better choose the correct composition.


Sat, 2019-08-03 08:12