

In [1]:

    
%load_ext autoreload
%autoreload 2



In [36]:

    
import warnings; warnings.simplefilter('ignore')
import matplotlib
import matplotlib.pyplot as plt
import rstoolbox
import pandas as pd
import seaborn as sns
import numpy as np
import copy
import os

sns.set(font_scale=1.5)

Scaffold Adaptability: BHRF1 Inhibitor

The aim of these experiments is to asses the ability to structurally mimic the designed inhibitor of the Epstein-Barr virus Apoptosis regulator BHRF1, a homolog of human Bcl-2 proteins that block apoptosis and is associated with cancer. The designed inhibitor's structure can be found in the PDB 4OYD; and is based on the same binding motif as the one found in PDB 2WH6 and uses as template for the rest of the scaffold the structure of a three helical domain as found in PDB 3LHP.
The key issue of this example is that the disposition of the 3 helix of the bundle need to change from 3LHP to 4OYD in order not to clash with BHRF1. Thus, we expect to show how, by generating the design on FFL in the presence of the binder, we will be able to promote this conformational change, showcasing the importance of this new feature in our protocol. This expected clash can be observed when superimposing BHRF1 to the 3LHP scaffold through the binding motif of 2WH6.

3LHP	4OYD

Measures

Due to the type of conformational change that we are trying to evaluate, global RMSD (i.e. the best possible RMSD obtained from all structural alignments of the two proteins) is not an optimal way to evaluate how well we capture the necessary structural change. Thus, during this analysis, we will also evaluate what we will call local RMSD. This measure derives from alingning the structures through the binding motif (that does not change during the FFL protocol) and evaluate the RMSD without re-doing the alignment. This measure captures with higher accuracy the change of correlative disposition between the secondary structures.

Labeling

Label	Definition
MOTIF	Residues belonging to the functional query motif
HOTSPOT	Residues of the motif that are considered non-editable
COLDSPOT	Residues of the motif that are considered editable
FLEXIBLE	Residues of the motif's terminals that are allowed to explore alternative conformations
TEMPLATE	Residues of the design that do NOT belong to the query motif
CONTEXT	Residues belonging to other proteins in the complex that are NOT the design chain

Method Variants

For all these tests, we have divided the FFL protocol into two parts:

The Folding: Managed by the NubInitioMover.
The RD cycle: The Relax/Design (RD) cycle tests different ways to generate the final desings from the re-folded poses.

classicFFL:

Label	PACK	DESIGN	BB	CHI
HOTSPOT	-	-	-	-
COLDSPOT	X	X	-	X
FLEXIBLE	X	-	X	X
TEMPLATE	X	X	X	X
CONTEXT	-	-	-	-

fullDesign / minimize:

Label	PACK	DESIGN	BB	CHI
HOTSPOT	-	-	X	X
COLDSPOT	X	X	X	X
FLEXIBLE	X	-	X	X
TEMPLATE	X	X	X	X
CONTEXT	-	-	X	X

(one uses FastRelax, the other uses MinMover)

minimizeBinderAlternate:

(alternates one round of classicFFL and one with the following profile)

Label	PACK	DESIGN	BB	CHI
HOTSPOT	-	-	-	-
COLDSPOT	-	-	-	-
FLEXIBLE	-	-	-	-
TEMPLATE	-	-	-	-
CONTEXT	-	-	X	X

A) Evaluating start-end variability

In order to do this comparisson, there are a couple of things we need to know:

What is the global and local RMSD between 3LHP (start) and 4OYD (end)
- Aligning the binding motif of 4OYD to the insertion region of 3LHP
- We call this tvt as template_vs_target.
How much does 4OYD (end) can naturally change depending on different minimization protocols
- This refers to apply the same R/D protocols (without sequence editing) that we will apply to our desings



In [3]:

    
tvt = rstoolbox.api.read_rosetta_silent("experiments/3lhp_vs_4oyd/3lhp_vs_4oyd")
definition = {
    "scores":{
        "description": "description", "GlobalRMSD": "GlobalRMSD", "LocalRMSD": "LocalRMSD",
        "LocalRMSDH": "LocalRMSDH", "LocalRMSDL": "LocalRMSDL",
    }
}
tvtDF = rstoolbox.api.process_from_definitions(tvt, definition)
tvtDF









    Out[3]:






  
    
      
      GlobalRMSD
      LocalRMSD
      LocalRMSDH
      LocalRMSDL
      description
    
  
  
    
      0
      1.605
      2.351
      2.197
      1.711
      3lhp_0001



In [5]:

    
experiments = ["classicFFL", "fullDesign", "minimize", "minimizeBinderAlternate"]
definition = {
    "scores":{
        "description": "description", "GlobalRMSD": "GlobalRMSD", "LocalRMSD": "LocalRMSD",
        "LocalRMSDH": "LocalRMSDH", "LocalRMSDL": "LocalRMSDL", "design_type": "experiment"
    }
}
logic = {
    "keep": ["description", "experiment", "GlobalRMSD", "condition", "test"],
    "split": [("LocalRMSD", "target"), ("LocalRMSDH", "helices"), ("LocalRMSDL", "corehelices") ],
    "names": ["rmsd", "rmsd_to"]
}
tmpDFL = []
for exp in experiments:
    tmp = rstoolbox.api.read_rosetta_silent("experiments/4oyd_moves/{0}".format(exp), exp)
    tmpDFL.append(rstoolbox.api.process_from_definitions(tmp, definition))
moves4oydDF = []
moves4oydDF.append(pd.concat(tmpDFL))
moves4oydDF[0] = moves4oydDF[0].assign(condition=pd.Series(["binder"]*len(moves4oydDF[0]["description"])).values)
moves4oydDF[0] = moves4oydDF[0].assign(test=pd.Series(["4oyd_moves"]*len(moves4oydDF[0]["description"])).values)
moves4oydDF.append(rstoolbox.api.split_columns(moves4oydDF[0], logic))



In [6]:

    
g = sns.FacetGrid(moves4oydDF[1], col="experiment", hue="rmsd_to", size=10, aspect=0.6, legend_out=True)
g = (g.map(sns.regplot, "GlobalRMSD", "rmsd", fit_reg=False).add_legend())
plt.subplots_adjust(top=0.9)
g.axes[0,0].set_ylim(0,3)
g.axes[0,0].set_xlim(0,3)
g.fig.suptitle('Global vs. Local RMSD for 4OYD movements')
sns.plt.show()

B) How close can we mimic 4OYD from itself?

Here we use 4OYD both as source of the query motif and binder and as source of the template. This means that there is no need for the backbone to reajust to fit the binding site.
For this experiment we only apply classicFFL.



In [7]:

    
conditions = ["nobinder", "binder"]
definition = {
    "scores":{
        "description": "description", "GRMSD2Target": "GlobalRMSD", "LRMSD2Target": "LocalRMSD",
        "LRMSDH2Target": "LocalRMSDH", "LRMSDLH2Target": "LocalRMSDL", "design_score": "score"
    },
    "naming": ["test", "", "condition", "", "cluster", "decoy"]
}
ddg_defs = {
    "scores":{
        "ddg": "pre_ddg", "post_ddg": "post_ddg", "rmsd_drift": "rmsd_drift"
    },
    "naming": ["test", "", "condition", "", "cluster", "decoy", ""]
}
logic1 = {
    "keep": ["description", "experiment", "GlobalRMSD", "condition", "test", "score" ],
    "split": [("LocalRMSD", "target"), ("LocalRMSDH", "helices"), ("LocalRMSDL", "corehelices") ],
    "names": ["rmsd", "rmsd_to"]
}
logic2 = {
    "keep": ["description", "experiment", "condition", "test", "score", "rmsd_drift" ],
    "split": [("post_ddg", "post-min"), ("pre_ddg", "pre-min") ],
    "names": ["ddg", "ddg_when"]
}
tmpDFL = []
ddgDFL = []
for cnd in conditions:
    tmp = rstoolbox.api.read_rosetta_silent("experiments/as4oyd/{0}/as4oyd_{0}_1_minisilent".format(cnd))
    tmpDFL.append(rstoolbox.api.process_from_definitions(tmp, definition))
    tmp = rstoolbox.api.read_rosetta_silent("experiments/as4oyd/{0}/ddg_evaluation.score".format(cnd))
    ddgDFL.append(rstoolbox.api.process_from_definitions(tmp, ddg_defs))
as4oydDF = []
as4oydDF.append(pd.concat(tmpDFL))
as4oydDF[0] = pd.merge(as4oydDF[0], pd.concat(ddgDFL), how='left', on=["test", "condition", "cluster", "decoy"])
as4oydDF[0] = as4oydDF[0].assign(experiment=pd.Series(["classicFFL"]*len(as4oydDF[0]["description"])).values)
as4oydDF.append(rstoolbox.api.split_columns(as4oydDF[0], logic1))
as4oydDF.append(rstoolbox.api.split_columns(as4oydDF[0], logic2))



In [8]:

    
g = sns.FacetGrid(as4oydDF[1], col="rmsd_to", hue="condition", size=10, aspect=0.6, legend_out=True)
g = (g.map(sns.regplot, "GlobalRMSD", "rmsd", fit_reg=False).add_legend())
plt.subplots_adjust(top=0.9)
g.axes[0,0].set_ylim(0,3)
g.axes[0,0].set_xlim(0,3)
g.fig.suptitle('Global vs. Local RMSD for 4OYD refolded on itself')
sns.plt.show()



In [9]:

    
g = sns.FacetGrid(as4oydDF[2], col="condition", hue="ddg_when", size=10, aspect=0.6, legend_out=True)
g = (g.map(sns.regplot, "ddg", "score", fit_reg=False).add_legend())
plt.subplots_adjust(top=0.9)
g.axes[0,0].set_xlim(-70,10)
g.fig.suptitle('Design Score vs. ddG 4OYD refolded on itself')
sns.plt.show()



In [35]:

    
f, axes = plt.subplots(2, 3, figsize=(20, 15))
sns.boxplot(x="condition", y="LocalRMSD", data=as4oydDF[0], showfliers=False, ax=axes[0,0])
sns.boxplot(x="condition", y="LocalRMSDH", data=as4oydDF[0], showfliers=False, ax=axes[0,1])
sns.boxplot(x="condition", y="LocalRMSDL", data=as4oydDF[0], showfliers=False, ax=axes[0,2])
sns.boxplot(x="condition", y="score", data=as4oydDF[0], showfliers=False, ax=axes[1,0])
sns.boxplot(x="condition", y="post_ddg", data=as4oydDF[0], showfliers=False, ax=axes[1,1])
sns.boxplot(x="condition", y="rmsd_drift", data=as4oydDF[0], showfliers=False, ax=axes[1,2])
plt.suptitle('4OYD refolded: RMSDs and ddG vs. score')

axes[0,0].set_ylim(0,3)
axes[0,1].set_ylim(0,3)
axes[0,2].set_ylim(0,3)
plt.show()

c) Use 4OYD as motif source and 3LHP as template



In [47]:

    
definition = {
    "scores":{
        "description": "description", "GRMSD2Target": "GlobalRMSD", "LRMSD2Target": "LocalRMSD",
        "LRMSDH2Target": "LocalRMSDH", "LRMSDLH2Target": "LocalRMSDL", "design_score": "score"
    },
    "naming": ["", "", "condition", "", "cluster", "decoy", "experiment", "", "", ""]
}
ddg_defs = {
    "scores":{
        "ddg": "pre_ddg", "post_ddg": "post_ddg", "rmsd_drift": "rmsd_drift"
    },
    "naming": ["", "", "condition", "", "cluster", "decoy", "experiment", "", "", "", ""]
}
logic1 = {
    "keep": ["description", "experiment", "GlobalRMSD", "condition", "test", "score" ],
    "split": [("LocalRMSD", "target"), ("LocalRMSDH", "helices"), ("LocalRMSDL", "corehelices") ],
    "names": ["rmsd", "rmsd_to"]
}
logic2 = {
    "keep": ["description", "experiment", "condition", "test", "score", "rmsd_drift" ],
    "split": [("post_ddg", "post-min"), ("pre_ddg", "pre-min") ],
    "names": ["ddg", "ddg_when"]
}
tmpDFL = []
ddgDFL = []
filename="experiments/from4oyd/{0}/{1}/design_{0}_1_minisilent"
ddgname="experiments/from4oyd/{0}/{1}/ddg_evaluation.score"
for exp in experiments:
    for cnd in conditions:
        if os.path.isfile(filename.format(exp, cnd)):
            tmp = rstoolbox.api.read_rosetta_silent(filename.format(exp, cnd))
            tmpDFL.append(rstoolbox.api.process_from_definitions(tmp, definition))
            tmp = rstoolbox.api.read_rosetta_silent(ddgname.format(exp, cnd), allow_repeats=True)
            ddgDFL.append(rstoolbox.api.process_from_definitions(tmp, ddg_defs))
from4oydDF = []
from4oydDF.append(pd.concat(tmpDFL))
from4oydDF[0] = pd.merge(from4oydDF[0], pd.concat(ddgDFL), how='left', on=["experiment", "condition", "cluster", "decoy"])
from4oydDF[0] = from4oydDF[0].assign(test=pd.Series(["from4oyd"]*len(from4oydDF[0]["description"])).values)
from4oydDF.append(rstoolbox.api.split_columns(from4oydDF[0], logic1))
from4oydDF.append(rstoolbox.api.split_columns(from4oydDF[0], logic2))



In [70]:

    
g = sns.FacetGrid(from4oydDF[1], col="rmsd_to", hue="condition", row="experiment", size=10, aspect=0.8, legend_out=True)
g = (g.map(sns.regplot, "GlobalRMSD", "rmsd", fit_reg=False).add_legend())
plt.subplots_adjust(top=0.9)
g.axes[0,0].set_ylim(0,5)
g.axes[0,0].set_xlim(0,5)
g.fig.suptitle('Global vs. Local RMSD for all from4OYD experiments')
sns.plt.show()



In [79]:

    
f, axes = plt.subplots(4, 4, figsize=(25, 20))
for i in range(0, 4):
    pltdf = from4oydDF[0][from4oydDF[0]["experiment"] == experiments[i]]
    sns.boxplot(x="condition", y="LocalRMSDL", data=pltdf, showfliers=False, ax=axes[i,0])
    axes[i,0].set_ylim(0,3.5)
    sns.boxplot(x="condition", y="score", data=pltdf, showfliers=False, ax=axes[i,1])
    axes[i,1].set_ylim(-350,-250)
    sns.boxplot(x="condition", y="post_ddg", data=pltdf, showfliers=False, ax=axes[i,2])
    axes[i,2].set_ylim(-70,0)
    sns.boxplot(x="condition", y="rmsd_drift", data=pltdf, showfliers=False, ax=axes[i,3])
    axes[i,3].set_ylim(0,12.5)
    
plt.suptitle('4OYD experiments: RMSDs and ddG vs. score')
plt.show()

Use 2WH6 as motif source and 3LHP as template



In [72]:

    
definition = {
    "scores":{
        "description": "description", "GRMSD2Target": "GlobalRMSD", "LRMSD2Target": "LocalRMSD",
        "LRMSDH2Target": "LocalRMSDH", "LRMSDLH2Target": "LocalRMSDL", "design_score": "score"
    },
    "naming": ["", "", "condition", "", "cluster", "decoy", "experiment", "", "", ""]
}
ddg_defs = {
    "scores":{
        "ddg": "pre_ddg", "post_ddg": "post_ddg", "rmsd_drift": "rmsd_drift"
    },
    "naming": ["", "", "condition", "", "cluster", "decoy", "experiment", "", "", "", ""]
}
logic1 = {
    "keep": ["description", "experiment", "GlobalRMSD", "condition", "test", "score" ],
    "split": [("LocalRMSD", "target"), ("LocalRMSDH", "helices"), ("LocalRMSDL", "corehelices") ],
    "names": ["rmsd", "rmsd_to"]
}
logic2 = {
    "keep": ["description", "experiment", "condition", "test", "score", "rmsd_drift" ],
    "split": [("post_ddg", "post-min"), ("pre_ddg", "pre-min") ],
    "names": ["ddg", "ddg_when"]
}
tmpDFL = []
ddgDFL = []
filename="experiments/from2wh6/{0}/{1}/design_{0}_1_minisilent"
ddgname="experiments/from2wh6/{0}/{1}/ddg_evaluation.score"
for exp in experiments:
    for cnd in conditions:
        if os.path.isfile(filename.format(exp, cnd)):
            tmp = rstoolbox.api.read_rosetta_silent(filename.format(exp, cnd))
            tmpDFL.append(rstoolbox.api.process_from_definitions(tmp, definition))
            tmp = rstoolbox.api.read_rosetta_silent(ddgname.format(exp, cnd), allow_repeats=True)
            ddgDFL.append(rstoolbox.api.process_from_definitions(tmp, ddg_defs))
from2wh6DF = []
from2wh6DF.append(pd.concat(tmpDFL))
from2wh6DF[0] = pd.merge(from2wh6DF[0], pd.concat(ddgDFL), how='left', on=["experiment", "condition", "cluster", "decoy"])
from2wh6DF[0] = from2wh6DF[0].assign(test=pd.Series(["from2wh6"]*len(from2wh6DF[0]["description"])).values)
from2wh6DF.append(rstoolbox.api.split_columns(from2wh6DF[0], logic1))
from2wh6DF.append(rstoolbox.api.split_columns(from2wh6DF[0], logic2))



In [73]:

    
g = sns.FacetGrid(from2wh6DF[1], col="rmsd_to", hue="condition", row="experiment", size=10, aspect=0.8, legend_out=True)
g = (g.map(sns.regplot, "GlobalRMSD", "rmsd", fit_reg=False).add_legend())
plt.subplots_adjust(top=0.9)
g.axes[0,0].set_ylim(0,5)
g.axes[0,0].set_xlim(0,5)
g.fig.suptitle('Global vs. Local RMSD for all from2WH6 experiments')
sns.plt.show()



In [82]:

    
f, axes = plt.subplots(4, 4, figsize=(25, 20))
for i in range(0, 4):
    pltdf = from2wh6DF[0][from2wh6DF[0]["experiment"] == experiments[i]]
    sns.boxplot(x="condition", y="LocalRMSDL", data=pltdf, showfliers=False, ax=axes[i,0])
    axes[i,0].set_ylim(0,9)
    sns.boxplot(x="condition", y="score", data=pltdf, showfliers=False, ax=axes[i,1])
    axes[i,1].set_ylim(-340,-200)
    sns.boxplot(x="condition", y="post_ddg", data=pltdf, showfliers=False, ax=axes[i,2])
    axes[i,2].set_ylim(-80,0)
    sns.boxplot(x="condition", y="rmsd_drift", data=pltdf, showfliers=False, ax=axes[i,3])
    axes[i,3].set_ylim(0,25)
    
plt.suptitle('2WH6 experiments: RMSDs and ddG vs. score')
plt.show()



In [ ]: