SPRINT 3.0:

Side-chain PRediction INference Toolbox

for Multistate Protein Design

SPRINT is released for academic research only under the General Public License. This package is free software as described in the GPLv3 license. For purposes other then academic research, please contact Menachem Fromer .

When using SPRINT, please cite:

Menachem Fromer, Chen Yanover, Amir Harel, Ori Shachar, Yair Weiss, and Michal Linial. SPRINT: side-chain prediction inference toolbox for multistate protein design. Bioinformatics, 26(19):2466-2467, 2010.

http://bioinformatics.oxfordjournals.org/content/26/19/2466.abstract

and:

Menachem Fromer, Chen Yanover, and Michal Linial. Design of multispecific protein sequences using probabilistic graphical modeling. Proteins: Structure, Function, and Bioinformatics, 78:530-547, 2010.

http://dx.doi.org/10.1002/prot.22575

SPRINT --- expanding the scope of protein design

SPRINT, acronym for Side-chain Prediction Inference Toolbox, is a package that uses inference algorithms on graphical models for the multistate protein design problem. It first represents the side-chain packing for a fixed-backbone protein design problem as a probabilistic graphical model. It can then use the dead-end_elimination (DEE) algorithm to reduce the dimensionality of the conformational spaces. Finally, it applies various inference algorithms on the resulting model.

SPRINT input:

Protein list input file for one or more 3D structures.
Each line in the protein list refers to a protein file, which refers to a rotamer energies file and rotamers file for that protein.
See the Examples for sample files.

NOTE: At this time, it is best to generate rotamer energy files using an external energy function (e.g., ORBIT, Rosetta, EGAD), though SPRINT does have in place a mechanism for generating energies starting from a PDB structure as input.

SPRINT output:

Multiple lowest-scoring amino acid sequences compatible with the target structures, in order of increasing energy score. This approach utilizes approximate max-product belief propagation (BP) together with the tBMMF algorithm, or the exact A* algorithm (when feasible).
Marginal amino acid probabilities corresponding to the full space of sequences compatible with the target structures. This utilizes sum-product BP.
Higher-order (e.g., pairwise) amino acid probabilities compatible with the target structures. This also uses sum-product BP.

Overview of SPRINT multistate design procedure:

The package is written in C++, and the probabilistic inference core of SPRINT is based on the fastInf package.

We believe that SPRINT's main advantage over other state-of-the-art protein design packages is its ability to perform varying functionalities better than other algorithms, while still using comparable memory and CPU resources.

Downloads

SPRINT's C++ source code and data files: sprint_3.0.tgz

Installation

To install SPRINT, run:: tar -xvzf sprint_3.0.tgz; cd sprint; make

NOTE 1: If you have any problems installing FastInf, please see the FastInf web site for known issues.

NOTE 2: The boost, glpk, and gsl packages are all included in the SPRINT package (under the sprint/sources/fastInf directory) so that FastInf functions properly.
If you are having problems installing these, you may need to download and install them instead of the versions included here.

NOTE 3: The GNU Gengetopt package must be installed before running the 'make' command for SPRINT installation. It can be downloaded from:
http://www.gnu.org/software/gengetopt/gengetopt.html#Installation

Usage

To run SPRINT, type:

build/execs/sprint [options]

SPRINT standard command-line options

-h, --help
Show the usage message and all standard options

-i, --inputFile
Specify a protein list input file for multistate design.
For side-chain prediction and/or single-state design ("--program=ssprint"), then a single protein file should be specified.

-n, --inference
Specify an inference task: finding lowest energy sequences ("MAPs" is the default), amino acid marginals ("marginals"), higher-order marginals ("cliq_marginals").
For side-chain prediction and/or single-state design ("--program=ssprint"), the conformational free energy ("free_energy") can also be specified.

-M, --Mconfigs
For MAPs inference, specify the number of sequences (or side-chain conformations, if "--program=ssprint")

-o, --energyOutputFile
Specify the name of the energies output file (otherwise, the default name is used)

-u, --proteinOutputFile
Specify the name of the protein output file (otherwise, the default name is used)

-t, --temperature
Specify the temperature (R * T) value used in the Boltzmann distribution (for "marginals", "cliq_marginals")

-s, --tool
Specify the inference tool: either loopy belief propagation ("loopy" is the default), generalized belief propagation ("GBP"), junction tree ("JT"), or A* search tree (A)

-D, --dee_threshold
Set the dead-end_elimination (DEE) threshold (otherwise, no DEE is performed).
DEE is performed on each protein individually as a pre-processing stage (before the inference performed with the user-specified "--tool").
Note that SPRINT performs type-dependent DEE for cases of multistate design ("--program=msprint"). However, with "--program=ssprint", then standard DEE is performed.

-g, --linksFile
Specify an amino acid-links file

--requireAllDesignCliqs
Specify that probabilities for all cliques of design variables in the problem are output.
For example, if specified in conjunction with "--maxCliqueVars=3", then probabilities will be output for ALL subsets of size 3.

--maxCliqueVars
Specify the size of design variable cliques to be output (e.g., "2" for pairwise)

-X, --multiStateBPMaxMessages
Specify the maximum number of messages to be iterated (for "loopy", "GBP", or "JT")

-p, --program
Choose to perform multi-state design ("msprint" is the default), OR side-chain prediction and/or single-state design ("ssprint").
In the case of "ssprint", the "--proteinOutputFile" output of SPRINT will contain atomic side-chain conformations (as rotamer dihedral angles).

--detailed-help
Show the usage message and all possible options (in full detail)

Examples

To run the following examples, download the relevant example file and unzip it:

Download example1.zip (6 MB)
To find the 10 lowest energy amino acid sequences compatible with the prion design problem (PDB code 1I4M) used in our single-state design paper:
```
sprint -i prion_list -n MAPs -M 10
```
The FASTA sequences are output to prion_list_design.seq, and their energies are listed in prion_list_design.en.actual.

Download example2.zip (27 MB)
To find the amino acid probabilities compatible with the PPAR design problem (PDB codes 1K74, 2PRG) used in our multi-state design paper, at a Boltzmann temperature of 0.41:
```
sprint -i PPAR_list -n marginals -t 0.41
```
The per-position marginal amino acid probabilities are output to PPAR_list_seq_marginals.prob.

Download example3.zip (5 MB)
To find the pairwise amino acid probabilities compatible with the design of the interface of the p53-53BP1 complex (PDB code 1KZY), at a Boltzmann temperature of 0.51:
```
sprint -i p53_list -n cliq_marginals --requireAllDesignCliqs --maxCliqueVars=2 -t 0.51
```
As specified by the use of the "--requireAllDesignCliqs --maxCliqueVars=2" flags, ALL pairwise amino acid probabilities, are output to p53_list_seq_cliq_marginals.prob.

Side-chain prediction and/or single-state design ("--program=ssprint") examples

Examples for running side-chain prediction and/or single-state design can be found in the previous version of SPRINT.

For Code Developers

Here is some automatically generated documentation for SPRINT C++ source code, including a graphical C++ class hierarchy.
NOTE that we already have in place a general and elaborate code mechanism for generating energies starting from a PDB structure as input (rather than protein files containing energies calculated by an external energy function).
Previous versions of SPRINT can be found here.