Author Archives: Konrad Krawczyk

Interesting Antibody Papers

Hints how broadly neutralizing antibodies arise (paper here). (Haynes lab here) Antibodies can be developed to bind virtually any antigen. There is a stark difference however between the ‘binding’ antibodies and ‘neutralizing’ antibodies. Binding antibodies are those that make contact with the antigen and perhaps flag it for elimination. This is in contrast to neutralizing antibodies, whose binding eliminates the biological activity of the antigen. A special class of such neutralizing antibodies are ‘broad neutralizing antibodies’. These are molecules which are capable of neutralizing multiple strains of the antigen. Such broadly neutralizing antibodies are very important in the fight against highly malleable diseases such as Influenza or HIV.

The process how such antibodies arise is still poorly understood. In the manuscript of Williams et al., they make a link between the memory and plasma B cells of broadly neutralizing antibodies and find their common ancestor. The common ancestor turned out to be auto-reactive, which might suggest that some degree of tolerance is necessary to allow for broadly neutralizing abs (‘hit a lot of targets fatally’). From a more engineering perspective, they create chimeras of the plasma and memory b cells and demonstrate that they are much more powerful in neutralizing HIV.

Ineresting data: their crystal structures are different broadly neutralizing abs co-crystallized with the same antigen (altought small…). Good set for ab-specific docking or epitope prediction — beyond the other case like that in the PDB (lysozyme)! At the time of writing the structures were still on hold in the PDB so watch this space…

Interesting Antibody Papers

Below are two somewhat recent papers that are quite relevant to those doing ab-engineering. The first one takes a look at antibodies as a collection — software which better estimates a diversity of an antibody repertoire. The second one looks at each residue in more detail — it maps the mutational landscape of an entire antibody, showing a possible modulating switch for VL-CL interface.

Estimating the diversity of an antibody repertoire. (Arnaout Lab) paper here. High Throughput Sequencing (or next generation sequencing…) of antibody repertoires allows us to get snapshots of the overall antibody population. Since the antibody population ‘diversity’ is key to their ability to find a binder to virtually any antigen, it is desirable to quantify how ‘diverse’ the sample is as a way to see how broad you need to cast the net. Firstly however, we need to know what we mean by ‘diversity’. One way of looking at it is akin to considering ‘species diversity’, studied extensively in ecology. For example, you estimate the ‘richness’ of species in a sample of 100 rabbits, 10 wolves and 20 sheep. Diversity measures such as Simpson’s index or entropy were used to calculate how biased the diversity is towards one species. Here the sample is quite biased towards rabbits, however if instead we had 10 rabbits, 10 wolves and 10 sheep, the ‘diversity’ would be quite uniform. Back to antibodies: it is desirable to know if a given species of an antibody is more represented than others or if one is very underrepresented. This might indicate healthy vs unhealthy immune system, indicate antibodies carrying out an immune response (when there is more of a type of antibody which is directing the immune response). Problem: in an arbitrary sample of antibody sequences/reads tell me how diverse they are. We should be able to do this by estimating the number of cell clones that gave rise to the antibodies (referred to as clonality). People have been doing this by grouping sequences by CDR3 similarity. For example, sequences with CDR3 identical or more than >95% identity, are treated as the same cell — which is tantamount to being the same ‘species’. However since the number of diverse B cells in a human organism is huge, HTS only provides a sample of these. Therefore some rarer clones might be underrepresented or missing altogether. To address this issue, Arnaout and Kaplinsky developed a methodology called Recon which estimates the antibody sample diversity. It is based on the expectation-maximization algorithm: given a list of species and their numbers, iterate adding parameters until they have a good agreement between the fitted distributions and the given data. They have validated this methodology firstly on the simulated data and then on the DeKosky dataset. The code is available from here subject to their license agreement.

Thorough analysis of the mutational landscape of the entire antibody. [here]. (Germaine Fuh from Affinta/Genentech/Roche). The authors aimed to see how malleable the variable antibody domains are to mutations by introducing all possible modifications at each site in an example antibody. As the subject molecule they have used high-affinity, very stable anti-VEGF antibody G6.31. They argue that this antibody is a good representative of human antibodies (commonly used genes Vh3, Vk1) and that its optimized CDRs might indicate well any beneficial distal mutations. They confirm that the positions most resistant to mutation are the core ones responsible for maintaining the structure of the molecule. Most notably here, they have identified that Kabat L83 position correlates with VL-CL packing. This position is most frequently a phenylalanine and less frequently valine or alanine. This residue is usually spatially close to isoleucine at position LC-106. They have defined two conformations of L83F — in and out:

  1. Out: -50<X1-100 interface.
  2. In: 50<X1<180

Being in either of these positions correlates with the orientation of LC-106 in the elbow region. This in turn affects how big the VL-CL interface is (large elbow angle=small  tight interface; small elbow angle=large interface). The L83 position often undergoes somatic hypermutation, as does the LC-106 with the most common mutation being valine.

Interesting Antibody Papers.

Below are several antibody papers that should be of interest to those dealing with antibody engineering, be it computational or experimental. The running motif in this post will be humanization, or the process of engineering a mouse antibody sequence which binds to a target to look ‘more human’ so as to reduce the immune response (if you need an early citation on this issue, here it is).

We present two papers which talk about antibody humanization directly, one from structural point of view (Choi et al. 2015), the other one highlighting issues facing antibody engineers  mining for information (Martin & Rees, 2016). The third paper (Collins et al. 2015) takes a step back from the issues presented in the other papers and talks broadly about the nature of mouse sequences raised in the lab.

Humanization via structural means [here] (Bailey-Kellogg group). The authors introduce a novel methodology named CoDAH to facilitate humanization of antibodies. They design an approach which makes a tradeoff between sequence and structural humanization scores. The sequence score used is the Human String Content (Laza et al. 2007, Mol Immunol), which calculates how similar the query (murine) sequence is to short stretches of human sequences (mostly germilne). In line with the fact that T-Cells are one of main drivers of anti-biologics immunity, they define the sequences stretches to be 9-mer, as recognized by T-Cells. For the structural score, they use Rotameric energy as calculated by Amber. They demonstrate that constructs designed using their score express and retain affinity towards the target antigen, however they do not appear to prove that the new sequences are not immunogenic.

Extracting data from databases for humanization [here] (Martin group and Rees consulting). The main purpose of this manuscript is to warn potential antibody engineers of the pitfalls of species mis-annotations. They point out that in a routine ‘humanization’ pipeline where we aim to find human sequences given a mouse sequence, a great number of seemingly good ‘human’ templates are not human at all (sources as diverse as IMGT or PDB). This might lead to errors down the line if the engineer does not double check the annotations (unfortunate but true). Many of such annotations arise because the cells in which mouse antibodies are expressed are human cells or because the sequences are chimeric — in either case the annotation would not read mouse or chimeric, but erroneously ‘human’. NB. Another thing to watch in this publication is the fact that authors are working on a sequence database of their own: EMBLIG which is said to collect data from EMBL-ENA (nucleotide repository from EMBL). Hopefully in their database, authors will address the issues that they point out here.

What can we say about antibodies produces by laboratory mice? [here] (Collins group). Authors of this manuscript have addressed the issue that the now available High Throughput Sequencing (HTS) overlooked mouse repertoires. Different mice strains have different susceptibilities to diseases (Houpt, 2002, J Imunol; which might mean that you need to think twice which mice strain to choose for a given target). Currently known antibody repertoire of mice is based on the sequencing of two strains, BALB/c and C57BL/6. Here the authors apply HTS to two strains (BALB/c and C57BL/6) of laboratory mice (eight mice per strain) to get a better snapshot of antibody gene usage. Specifically, they pay close attention to the different genes combinations (VDJ) in the sequences that they obtain. Authors conclude that the repertoires between the two strains are strikingly different and quite restricted — which might mean that the laboratory mice were under very specific pressures (read inbred/overbred). All in all, the VDJ usage numbers that they produce in this publication are a useful reference to know which sequence combinations might be used by antibody engineers.

Interesting Antibody Papers

De Novo H3 prediction by C-terminal kink-biasing (Gray Lab) [here].

Authors introduce an improvement to the prediction of CDR-H3 in the form of a constraint for de-novo decoy generation. Working from the observation that 80% of CDR-H3 have kinked C-Terminal (Weitzner et al., 2015, Structure), they bias the loops to assume this conformation (they prove that it does not force ALL loops to do so!). The constraint is in the form of a pseudo bond angle between Ca for the three C-terminal residues and a pseudo dihedral angle for the three C-terminal residues and one adjacent residue in the framework. The bias takes the form of a penalty score if the generated angle falls outside mean +/- 1s. They use a quite stringent H3 loop benchmark of only 49 loops. Using this constraint on this dataset improves prediction for majority of the loops. They also demonstrate the utility of the score for full Fv homology modeling and Ab-Ag docking.

Therapeutic vs synthetic vs natural antibodies (Ofran Lab) [here].

The authors analyzed 137 Ab-Ag complexes from the PDB. Those from hybridoma and synthetic libraries were classified as ‘Natural’ and those coming from ‘synthetic’ libraries. They demonstrate that synthetic libraries overuse H3 in the number of contacts the antibody forms with the antigen, whereas natural constructs share the paratope with H1& H2 to a larger extent. This, together with their tool, CDRs analyzer (analysis of structural & biochemical properties of ab-ag complex) can be a useful method to inform the design of antibodies.

From the past: TABHU, tools for antibody humanization (Tramontano Lab) [here]. Authors have created a tool to aid antibody humanization. Given a sequence of an antibody, the system would look for the most suitable template from their extensive sequence databases (DIGIT) and germline sequences from IMGT. The templates are assessed on sequence similarity to the query and the similarity of the ‘binding’ mode which is assessed by their paratope prediction tool proABC. After the template had been chosen, the user can produce a structural model of the sequence.

What happens to the Human Immune Repertoire over time?

Last week during the group meeting we talked about a pre-print publication from the Ippolito group in Austin TX (here). The authors were monitoring the antibody repertoire from Bone Marrow Plasma Cells (80% of circulating abs) over a period of 6.5 years. For comparison they have monitored another individual over the period of 2.3 years. In a nutshell, the paper is like Picture 1 — just with antibodies

This is what the paper talks about in a nutshell. How the antibody repertoire looks like taken at different timepoints in an individual's lifetime.

This is what the paper talks about in a nutshell. How the antibody repertoire looks like taken at different timepoints in an individual’s lifetime.


The main question that they aimed to answer was: ‘Is the Human Antibody repertoire stable over time‘? It is plausible to think that there should be some ‘ground distribution’ of antibodies that are present over time which act as a default safety net. However we know that the antibody makeup can change radically especially when challenged by antigen. Therefore, it is interesting to ask, does the immune repertoire maintain a fairly stable distribution or not?

Firstly, it is necessary to define what we consider a stable distribution of the human antibody repertoire. The antibodies undergo the VDJ recombination as well as Somatic Hypermutation, meaning that the >10^10 estimated antibodies that a human is capable of producing have a very wide possible variation. In this publication the authors mostly focused on addressing this question by looking at how the usage of possible V, D and J genes and their combinations changes over time.

Seven snapshots of the immune repertoire were taken from the individual monitored over 6.5 years and two from the individual monitored over 2.3 years. Looking at the usage of the V, D and J genes over time, it appears that the proportion in each of the seven time points appears quite stable (Pic 2). Authors claim similar result looking at the combinations.  This would suggest that our antibody repertoire is biased to sample ‘similar’ antibodies over time. These frequencies were compared to the individual who was sampled over the period of 2.3 years and it appears that the differences might not be great between the two.

How the frequencies of V, D  and J genes change (not) over 6.5 years in a single individual

How the frequencies of V, D and J genes change (not) over 6.5 years in a single individual

It is a very interesting study which hints that we (humans) might be sampling the antibodies from a biased distribution — meaning that our bodies might have developed a well-defined safety net which is capable of raising an antibody towards an arbitrary antigen. It is an interesting starting point and to further check this hypothesis, it would be necessary to carry out such a study on multiple individuals (as a minimum to see if there are really no differences between us — which would at the same time hint that the repertoire do not change over time).


Convergent affinity maturation.

Antibodies are the first line of defense of our organisms against noxious substances. They are the proteins which we ‘train’ to recognize noxious substances when we get immunized. Therefore understanding the immune response after being presented with an antigen is instrumental in developing novel vaccines.

One hypothesis relating to immune response to an antigen is that different organisms are likely to raise similar or even identical antibodies against the same antigen. Testing this hypothesis has become more realistic recently with the advent of Next Generation Sequencing technologies (NGS). Using NGS techniques it is possible to interrogate the sequential makeup of a large set of B-cells.

Such a study was conducted not that long time ago by Trueck et al. They have analysed antibody repertoires of five individuals, pre and post immunization to check if the immune systems converged on similar antibody sequences. The five individuals were immunized with a conjugate vaccine of HiB, MenC and TT. The antibodies were sequenced from cells extracted pre-vaccination and seven days after vaccination.

Firstly, the antibody repertoire appeared to reflect the fact that an organism was mounting an immune response as the clonality post vaccination was higher than before vaccination (more cells producing similar antibodies). Secondly, authors focused on identifying sequences from the public repertoire — those antibodies that are shared between individuals. This analysis focused on the CDR3 only, of which 47 were shared between at least two of the five individuals. Quite a large proportion of those sequences were known to be specific towards HiB and the enrichment of these was muhch higher in the post-vaccination sample. Only one sequence in this set was known previously to target TT. Nevertheless, relaxing the sequence similarity condition, a lot of sequences related to those known to be TT-specific were found among the five individuals. Most importantly, the number of such sequences was much higher in the post-vaccination samples, indicating that these might indeed have been raised in response to TT stimulation. The same was not true for MenC as hardly any sequences related to this antigen were found in the immune response of the five individuals.

Therefore, authors claim that looking at the enrichment of such shared sequences can be an indicator of the effectiveness of the immune response. They correlate statistics coming from looking at the number of shared sequences which appear to have moderate correlation to the antibody avidity data (even though p-values in some cases are quite high). This indicates that even in such a small set of individuals, antibodies are capable of converging on similar solutions. This might provide clues as to the characteristics antibodies that recognize specific antigens and thus facilitate novel vaccine design.




Drawing Custom Unrooted Trees from Sequence Alignments

Multiple Sequence Alignments can provide a lot of information relating to the relationships between proteins. One notable example was the map of the kinome space published in 2002 (Figure 1).


Figure 1. Kinase space as presented by Manning et al. 2002;

Such images organize our thinking about the possible space of such proteins/genes going beyond long lists of multiple sequence alignments. The image in Figure 1, got a revamp later which now is the popular ‘kinome poster’ (Figure 2).

Revamped dendrogram of the kinome fro Fig. 1. Downloaded from

Here we have created a script to produce similar dendrograms straight from the multiple sequence alignment files (although clearly not as pretty as Fig 2!). It is not difficult to find software that would produce ‘a dendrogram’ from an MSA but making it do the simple thing of annotating the nodes with colors, shapes etc. with respect to the labels of the genes/sequences is slightly more problematic. Sizes might correspond to the importance of given nodes and colors can organize by their tree branches. The script uses the Biopython module Phylo to construct a tree from an arbitrary MSA and networkx to draw it:

import networkx, pylab
from networkx.drawing.nx_agraph import graphviz_layout
from Bio import Phylo
from Bio.Phylo.TreeConstruction import DistanceCalculator
from Bio.Phylo.TreeConstruction import DistanceTreeConstructor
from Bio import AlignIO

#What color to give to the edges?
e_color = '#ccccff'
#What colors to give to the nodes with similar labels?
color_scheme = {'RSK':'#e60000','SGK':'#ffff00','PKC':'#32cd32','DMPK':'#e600e6','NDR':'#3366ff','GRK':'#8080ff','PKA':'magenta','MAST':'green','YANK':'pink'}
#What sizes to give to the nodes with similar labels?
size_scheme = {'RSK':200,'SGK':150,'PKC':350,'DMPK':400,'NDR':280,'GRK':370,'PKA':325,'MAST':40,'YANK':200}

#Edit this to produce a custom label to color mapping
def label_colors(label):
	color_to_set = 'blue'
	for label_subname in color_scheme:
		if label_subname in label:
			color_to_set = color_scheme[label_subname]
	return color_to_set

#Edit this to produce a custom label to size mapping
def label_sizes(label):
	#Default size
	size_to_set = 20
	for label_subname in size_scheme:
		if label_subname in label:
			size_to_set = size_scheme[label_subname]
	return size_to_set

#Draw a tree whose alignment is stored in msa.phy
def draw_tree():
	#This loads the default kinase alignment that should be in the same directory as this script
	aln ='agc.aln', 'clustal')
	#This will construct the unrooted tree.
	calculator = DistanceCalculator('identity')
	dm = calculator.get_distance(aln)
	constructor = DistanceTreeConstructor()
	tree = constructor.nj(dm)
	G = Phylo.to_networkx(tree)
	node_sizes = []
	labels = {}
	node_colors = []
	for n in G:
		label = str(n)
		if 'Inner' in label:
			#These are the inner tree nodes -- leave them blank and with very small sizes.
			node_sizes.append( 1 )
			labels[n] = ''
			#Size of the node depends on the labels!
			node_sizes.append( label_sizes(label) )
			#Set colors depending on our color scheme and label names
			#set the label that will appear in each node			
			labels[n] = label
	#Draw the tree given the info we provided!
	pos = graphviz_layout(G)
	networkx.draw(G, pos,edge_color=e_color,node_size = node_sizes, labels=labels, with_labels=True,node_color=node_colors)
	#Saving the image -- uncomment

if __name__ == '__main__':

We are going to use the kinase alignment example to demonstrate how the script can be used. The kinase alignment we use can be found here on the website. We load the alignment and construct the unrooted tree using the Bio.Phylo module. Note that on each line of the alignment there is a name. These names are the labels that we use to define the colors and sizes of nodes. There are two dummy functions that achieve that label_nodes() and label_sizes() — if you look at them it should be clear how to define your own custom labeling.

If you download the code and the alignment and run it by:


You should see a similar image as in Fig 3.

Fig 3. Size-color-customized unrooted tree straight from a multiple sequence alignment file of protein kinases. Constructed using the script








Journal Club: The Origin of CDR H3 Structural Diversity

Antibody binding site is broadly composed of the six hypervariable loops, the CDRs. There are three loops on the antibody light chain (L1, L2 and L3) and three loops on the antibody heavy chain (H1, H2 and H3).

Out of the six loops, five appear to adopt a constrained set of structural conformations (L1, L2, L3, H1 and H2). The conformations of H3 appear much less constrained, which was suggested to be the result of its higher relative importance in antigen recognition (however it is not a necessary condition). The only observations to date about the shapes of CDR-H3 is the existence of the extended and kinked conformations of its anchor.

The function of the kink was investigated recently by Weitzner et al. Here, the authors contrasted the geometry found in the antibody CDR-H3 loops to a set of 15k non-antibody polypeptides. They found that even though the extended conformation appears to be more favorable, the kinked one can also be found in many cases, particularly in the PDZ domains.

Weitzner et al. find that the extended conformation is much more common in non-antibody loops. However, the kinked conformation, even though less frequent is not outright rare. The situation is the opposite in antibodies where the majority of H3 conformations are kinked rather than extended.

The authors contrasted the sequence patterns of kinked antibody loops and kinked non-antibody loops and did not find anything predictive of the kinked conformation — suggesting that the effect might be non-local. Nonetheless, the secondary structure pattern of the kinked H3 and the kinked non-antibody loops appears similar.

Even though there might be no sequence-kink link, the authors indicate how their findings might improve H3 structure prediction. They demonstrate that admixing the kinked non-antibody loops into a template dataset for an H3 modeling software might provide more relevant templates.

In conclusion, the main message of the paper (selon moi) is putting forward of the hypothesis as to the role of the H3 kink. Since the kink is much more prevalent in H3 than in non-antibody proteins, there is a strong suggestion that there might be a special role for it. The authors suggest that the kinked conformation allows for more structural diversity, that would otherwise be restricted in the more rigid beta-stranded extended conformation. Thus, antibodies might have opted for a system wherein, they do not need to add dramatic mutations to their H3 in order to get more structural flexibility.


Graphical User Interface for MOSAICS as a Pymol plugin — PymoSAICS

MOSAICS is suite of sampling methods for molecular simulations of motion of nucleic acid and protein structures. It’s applicability has been demonstrated in simulating large ensembles of nucleic acids (Sim 2012, Minary 2014) and proteins.

Starting with a protein/dna/rna structure you would like to examine, the basic modus operandi of MOSAICS is divided into three parts:

  • Pick your energy function or a statistical/empirical potentials (e.g. empirical Amber or CHARMM)
  • Pick your sampling methodology — e.g. parallel tempering
  • Details of simulation: solvent (implicit?), degrees of freedom (cartesian, torsional?) etc.

The energy function defines the energy surface with respect to your degrees of freedom (DoFs) and the sampling methodology is supposed to explore the conformational space along DoFs.

One of the main fortes of MOSAICS lies in the ability of defining hierarchichal natural moves. Defining regions of collective motion introduces experimental knowledge and intuition into the simulations, greatly accelerating sampling. Ability to define such regions  was one of the main reasons to start the development of the graphical user interface (GUI) for MOSAICS — preliminary version can be seen in the Figure below.


Overview of PymoSAICS in its current form.

Since we are developing the GUI as a plugin for Pymol, we called it PymoSAICS. The initial focus of the project is on nucleic acids due to our interests in the structural effects of epigenetic modifications. As demonstrated in the Figure above, the GUI is divided into three main panels:

  • Current run — prepare a simulation
  • Simulation Manager — manage previous runs, import, export protocols
  • Help — That’s just a link to our website!

Users can upload their favorite structure via PymoSAICS or Pymol and play with the available parameters. The GUI is also an ongoing effort to streamline the available protocols in MOSAICS to shield the user from the many parameters that are available but perhaps not relevant to the simulation at hand.

We are currently starting beta tests of the application which (if you don’t mind not getting any support just yet) is available here, Therefore, if you are interested in becoming a tester please let me know, and you will receive a version around Easter (April-ish)! Contact me via konrad.krawczyk at


Antibody modeling via AMA II and RosettaAntibody


Protein modeling is one of the most challenging problems in bioinformatics. We still lack a clear theoretical framework which would allow us to link linear protein sequence to its native 3D coordinates. Given that we only have the structures for about a promile of the known seqs, homology modeling is still one of the most successful methods to obtain a structure from a sequence. Currently, using homology modeling and the 1393 known folds we can produce models for more than half known domains. In many cases this is good enough to get an overall idea of the fold but for actual therapeutic applications, there is still a need for high-resolution modeling.

There is one group of molecules whose properties can be readily exploited via computational approaches for therapeutic applications: antibodies.  With blockbuster drugs such as Humira, Avastin or Remicade, they are the leading class of biopharmaceuticals. Antibodies share a great degree of similarity with one another (<50-60% sequence identity) and there are at least 1865 antibody structures in the PDB. Therefore, homology modeling of these structures at high resolution becomes tractable, as exemplified by WAM and PIGS. Here, we will review the antibody modeling paradigm using one of the most successful antibody modeling tools, RosettaAntibody, concluding with the most recent progress from AMA II (antibody CASP).

General Antibody-antigen modeling

Modeling of antibody structures can be divided into the following steps:

  1. Identification of the Framework template
  2. Optimizing Vh/Vl orientation of the template
  3. Modeling of the non-H3 CDRs
  4. Modeling of H3

Most of the diversity of antibodies can be found in the CDRs. Therefore, the bulk of the protein can be readily copied from the framework region. This however needs to undergo an optimization of the Vh/Vl orientation. Prediction of the CDRs is more complicated since they are much more variable than the rest of the protein. Non-H3 CDRs can be modeled using canonical structure paradigms. Prediction of H3 is much more difficult since it does not appear to follow the canonical rules.

When the entire structure is assembled, it is recommended to perform refinement using some sort of relaxation of the structure, coupled with an energy function which should guide it.


RosettaAntibody protocol roughly follows this described above. In the first instance, an appropriate template is identified by highest BLAST bit scores. The best heavy and light chains aligned to the best-BLAST-scoring Fv region. The knowledge-base here is a set of 569 antibody structures form SACS with resolutions 3.5A and better. The Vh/Vl orientation is subsequently refined using local relaxation, guided by Charmm.

Non-H3 CDRs are modeled using the highest-scoring BLAST hit of the same length. Canonical information is not taken into account. Loops are grafted on the framework using the residues overlapping with the anchors.

H3 loops are modeled using a fragment based approach. The fragment library is Rosetta+H3 from the knowledge base of antibody structures created for the purpose of this study. The low-resolution search consists of Monte Carlo attempts to fit 3-residue fragments followed by Cyclic Coordinate Descent loop closure. This is followed by high resolution search when the H3 loop and Vh/Vl are repacked using a variety of moves.

Each decoy coming from the repacking is scored using Rosetta function. The lower the Rosetta score the better the decoy (according to Rosetta).


RosettaAntibody can produce high-quality models (1.4A) on its 54 structure benchmark test. The major limitation of the method (just like any other antibody modeling method) is the H3 loop modeling. It is believed that H3 is the most important loop and therefore getting this loop right is a major challenge.

Right framework and the correct orientation of Vh/Vl have a great effect on the quality of H3 predictions. When the H3 was modeled on using the correct framework, the predictions are order of magnitude better than by using the homology model. This was demonstrated using the native recovery in RosettaAntibody study as well as during ‘Step II’ of the Antibody Modeling assessment where participants were asked to model H3 using the correct framework.