UCSF Chimera--a visualization system for exploratory research and analysis.

doi:10.1002/JCC.20084

UCSF Chimera—A Visualization System for Exploratory

Research and Analysis

ERIC F. PETTERSEN, THOMAS D. GODDARD, CONRAD C. HUANG, GREGORY S. COUCH,

DANIEL M. GREENBLATT, ELAINE C. MENG, THOMAS E. FERRIN

Computer Graphics Laboratory, Department of Pharmaceutical Chemistry, University of

California, 600 16th Street, San Francisco, California 94143-2240

Received 24 February 2004; Accepted 6 May 2004

DOI 10.1002/jcc.20084

Published online in Wiley InterScience (www.interscience.wiley.com).

Abstract: The design, implementation, and capabilities of an extensible visualization system, UCSF Chimera, are

discussed. Chimera is segmented into a core that provides basic services and visualization, and extensions that provide

most higher level functionality. This architecture ensures that the extension mechanism satisﬁes the demands of outside

developers who wish to incorporate new features. Two unusual extensions are presented: Multiscale, which adds the

ability to visualize large-scale molecular assemblies such as viral coats, and Collaboratory, which allows researchers to

share a Chimera session interactively despite being at separate locales. Other extensions include Multalign Viewer, for

showing multiple sequence alignments and associated structures; ViewDock, for screening docked ligand orientations;

Movie, for replaying molecular dynamics trajectories; and Volume Viewer, for display and analysis of volumetric data.

A discussion of the usage of Chimera in real-world situations is given, along with anticipated future directions. Chimera

includes full user documentation, is free to academic and nonproﬁt users, and is available for Microsoft Windows,

Linux, Apple Mac OS X, SGI IRIX, and HP Tru64 Unix from http://www.cgl.ucsf.edu/chimera/.

Key words: molecular graphics; extensibility; visualization; multiscale modeling; sequence alignment

Introduction

Since its inception, the UCSF Computer Graphics Laboratory has

worked on molecular visualization systems to meet the needs of

researchers in the ﬁeld, beginning with MMS/MIDS

1

in 1976 to

our current offering, UCSF Chimera

2

(henceforth “Chimera”).

Chimera’s immediate predecessor, the Midas/MidasPlus system

3

(henceforth “Midas”), provided us with the insight that extensibil-

ity should be considered critically important in the design of a

visualization system.

Midas was a highly successful molecular graphics system.

However, it was relatively difﬁcult for users to add new function-

ality. The ﬁrst extension mechanism introduced into Midas was the

ability to send an annotated Protein Data Bank (PDB)

4

ﬁle repre

-

senting the currently displayed scene to an external program.

5

This

was motivated by our desire to interface to rendering programs

such as RASTER3D.

6

Once this extension mechanism was avail

-

able, it became relatively easy to develop rendering “back ends” of

our own, and soon thereafter we developed both a fast space-ﬁlling

renderer with shadows, Conic,

7

and a Jane Richardson-style

8

rib

-

bon-depiction program with many capabilities, Ribbonjr. Further-

more, outside developers exploited the mechanism in ways that we

had not anticipated. Thomas Hynes (then with Genentech) wrote

Neon, a program to allow Midas to depict shadowed ball-and-stick

scenes. Neon acted as a ﬁlter between Midas and Conic, taking the

PDB ﬁle output by Midas and producing another PDB ﬁle with an

order of magnitude more atoms. Neon output contained larger-

radius atoms for balls and series of many closely spaced, smaller-

radius atoms to simulate sticks. To our surprise, Conic was able to

process the Neon ﬁle quickly enough (seconds to minutes on

computers of the day) to be usable. This whole Neon concept was

something that would never have occurred to us, and opened our

eyes to the power of an effective extension mechanism.

A second extension mechanism was subsequently created to

allow Midas to communicate on an ongoing basis with external

programs. The Midas user could send commands to an external

program and the external program could issue Midas commands to

cause changes in the Midas session. Although this mechanism was

also quite successful, it was ultimately constrained to the available

Midas command set. The restrictions imposed by the original

Correspondence to: T.E. Ferrin; e-mail: tef@cgl.ucsf.edu

Contract/grant sponsor: NIH; contract/grant number: P41-RR01081

Midas design were proving problematic, and we were motivated to

create a new system with greater extensibility.

Chimera was designed with extensibility as a primary goal. We

also wanted Chimera to be portable to a wide variety of platforms,

and to include state-of-the-art graphics capabilities such as trans-

parency and interactive ball-and-stick, space-ﬁlling, ribbon, and

solid surface representations. Another design goal was to make

Chimera accessible to users at all skill levels by providing both a

graphical menu/window interface and a command-line interface.

Chimera Architecture

Chimera’s primary programming language is Python.

9

Impor

-

tantly, Python is an interpreted, object-oriented programming lan-

guage that is also easy to learn and very readable. Because Python

is interpreted, it is good for rapid development and debugging.

Readability is important for a team development project like

Chimera, and an easy-to-learn language enables others to develop

extensions without undue effort. Chimera includes the Python-

standard IDLE interactive development environment

10

to help

diagnose problems during extension development.

Chimera is divided into a core and extensions. The core pro-

vides basic services and molecular graphics capabilities. All higher

level functionality is provided through extensions. This design,

with the bulk of Chimera functions provided by extensions, en-

sures that the extension mechanism is robust enough to handle the

needs of outside researchers wanting to extend Chimera in novel

ways. Extensions can be integrated into the Chimera menu system,

and can present a separate graphical user interface as needed using

the Tkinter,

11

Tix,

12

and/or Pmw

13

toolkits.

The Chimera core consists of a C⫹⫹ layer that handles time-

critical operations (e.g., graphics rendering) and a Python layer

that handles all other functions. All signiﬁcant C⫹⫹ data and

functions are made accessible to the Python layer. Core capabili-

ties include molecular ﬁle input/output, molecular surface gener-

ation using the MSMS algorithm,

14

and aspects of graphical dis

-

play such as wire-frame, ball-and-stick, ribbon, and sphere

representations, transparency control, near and far clipping planes,

and lenses (screen areas with different display attributes; see

Fig. 1).

Another core service is maintenance and display of the current

selection. Users may select parts of structures by picking with the

mouse, by making menu choices (e.g., aromatic rings), or via

certain extension actions. The selected structure areas are high-

lighted with a particular color or a colored outline. Extensions can

query for the contents of the selection. Many menu actions (such

as coloring or setting the display style) work on the current

selection.

Figure 1. A single frame of a molecular dynamics trajectory of a

buckytube in water, shown with the Movie extension. A “lens” has

been placed over the center of the tube to strip away the surface and

reveal the hydrogen-bonded chain of waters passing through the tube.

The trajectory was computed with polarizable molecular dynamics

using the Amber 8

26

Sander module (J. Caldwell, UCSF, unpublished

data).

Figure 2. Bluetongue virus core particle (PDB identiﬁer 2btv

16

) with

double-stranded RNA attached to the surface (1h1k

17

) (top). Trimers

in the outer protein layer that are equivalent under the icosahedral

symmetry are given the same color. The free end of the RNA attaches

to other viral particles in the crystal. A closeup of the inner layer

(bottom) shows ball-and-stick and ribbon models and a surface at

higher resolution.

1606 Pettersen et al. • Vol. 25, No. 13 • Journal of Computational Chemistry

The core also maintains a trigger mechanism wherein changes

to core data structures or state are reported to extensions that have

registered callbacks with the corresponding trigger. For example,

there is a “selection changed” trigger that ﬁres whenever the

current selection changes.

Extensions are written either entirely in Python or in a combi-

nation of Python and C/C⫹⫹ (the latter using a shared library

loaded at runtime). Extensions can be placed in the Chimera

installation directory (which would make the extension available

to all users) or in the user’s own ﬁle area. Extensions are loaded on

demand, typically when the user accesses a menu entry that starts

the extension. The class structure of molecular data and other

extension programming information can be found at http://www.

cgl.ucsf.edu/chimera/docs/ProgrammersGuide/Examples/.

We demonstrate Chimera’s extensibility by presenting several

extensions. The Multiscale and Collaboratory extensions are quite

unique, and demonstrate the wide spectrum of abilities that can be

generated. The others provide insight into the use of core facilities

by extensions and the integration of extensions with the Chimera

environment.

Multiscale

The Multiscale extension adds capabilities for interactively explor-

ing large molecular assemblies. We have focused on viral struc-

tures, condensed chromosomes, and ribosomes; additional exam-

ples include cytoskeletal ﬁbers and motors, ﬂagellar structures, and

chaperonins. The Multiscale extension displays structures from the

PDB

4

and generates their multimeric forms by using transforma

-

tion matrices to position the subunits. Multiscale can also be used

with large assemblies where there are no repeated subunits. PDB

chains can be displayed as low-resolution surfaces, or in any of the

standard molecular representations available in Chimera. Multi-

scale permits biologically meaningful levels of quaternary struc-

ture to be deﬁned. The abilities to build multimeric forms, display

low-resolution representations, and deﬁne levels of structure are

important for interactively exploring large complexes.

Multiscale uses Chimera’s core molecular display abilities, data

structures, ﬁle reading, and selection management, and the Volume

Viewer extension for surface calculation and rendering. This

ready-to-use infrastructure allowed us to focus on the new capa-

bilities needed for displaying complexes.

Most of the available atomic-resolution viral structures are

icosahedral particles with 60-fold symmetry. PDB ﬁles provide

atomic coordinates for only one subunit. To build a multimeric

model, subunits are positioned using rotation/translation matrices

read from the PDB ﬁle header. PDB “REMARK 350” records give

matrices that can be used to generate the biological oligomeriza-

tion state. Crystallographic (SMTRY records) and noncrystallo-

graphic symmetry (MTRIX records) matrices or matrices inferred

from the space group (CRYST1 record) of crystal structures can

also be used. Multiscale’s visualization capabilities have revealed

shortcomings in the matrix information for many large-scale struc-

ture entries that otherwise would have been difﬁcult to detect. We

are working with the PDB to ﬁnd and correct these entries.

For efﬁciency, the Multiscale extension only loads atomic

coordinates for subunits when they are needed. When a model is

ﬁrst displayed, only a low-resolution surface representation is

shown, so no additional copies of the coordinates need to be

loaded. Chimera’s atomic and residue-level display styles are also

available, but are typically used for only a small number of

subunits so that the amount of detail depicted does not overwhelm

the user. The Multiscale extension does not use the high-resolution

molecular surfaces that are a core feature of Chimera. Such sur-

faces would render too slowly for a large multimer and provide too

high a level of detail to best illustrate the organization of subunits.

The Multiscale extension was originally written entirely in

Python. To speed up the surface calculation, certain critical rou-

tines were rewritten in C⫹⫹. Converting these routines from

interpreted Python to compiled C⫹⫹ made them run about 50

times faster. Translation is generally straightforward because Py-

thon objects such as molecules, atoms, and lists have equivalent

C⫹⫹ objects. Rendering the surface with OpenGL,

15

another

time-critical step, uses the C⫹⫹ module in the Volume Viewer

extension.

Subunits can be selected with the mouse. To simplify selecting

larger pieces of a structure, new structure levels can be deﬁned

hierarchically. For example, the bluetongue virus core particle

16

has two protein layers, the outer layer being composed of 260

trimers (Fig. 2). For this structure, it is useful to deﬁne inner and

outer layers and trimers as structural levels. After an individual

outer layer monomer is selected, the selection can then be pro-

moted to the containing trimer, and subsequently promoted to the

whole layer of trimers. The whole outer layer can then be hidden

if the object of interest is the inner protein layer. Besides being

promoted, a selection can be extended to all identical copies of the

currently selected subunits. It is also possible to select just the

subunits for which atomic information has been loaded.

Structure levels can be speciﬁed in a Python script. Structural

hierarchies are sometimes described in text in the headers of PDB

ﬁles. The mmCIF ﬁle format available from the PDB

4

has a limited

ability to describe such higher levels of structure in a computer-

readable form, but few submitted data sets provide this informa-

tion.

For multimeric structures, investigations are often facilitated by

the presence of more than one copy of the asymmetric unit. The

bluetongue virus structure

16,17

illustrates how working with the

full viral shell can aid analysis. The crystallographic data used to

determine the capsid structure revealed viral double-stranded RNA

stuck to the outside of the particles

17

(Fig. 2). To investigate the

speciﬁc atomic contacts between the RNA and virus, it is helpful

to locate the several subunits of the icosahedral particle adjacent to

the RNA by inspecting the full particle. These can then be exam-

ined using an all-atom display to determine the contacts account-

ing for the stickiness of the capsid.

The Multiscale extension is intended for problems where both

large-scale and atomic-scale details are relevant. The tools needed

to explore models with many levels of structure and large numbers

of atoms are necessarily complex; the Chimera Multiscale exten-

sion has addressed only the most immediate needs. We anticipate

increasing its capabilities signiﬁcantly in future releases.

Multalign Viewer

The Multalign Viewer extension allows Chimera to display se-

quence alignments together with associated structures (Fig. 3).

UCSF Chimera 1607

Multalign Viewer can read and write sequence alignments in a

wide variety of popular formats (currently Clustal

18

ALN,

“aligned” FASTA, GCG MSF, GCG RSF, “aligned” NBRF/PIR,

and Stockholm).

Multalign Viewer facilitates analysis of alignments in the con-

text of structural information and vice versa. First, structures in

Chimera must be associated with their corresponding sequences in

an alignment. When Multalign Viewer opens an alignment, it

examines the structures currently open in Chimera and checks each

chain for high sequence identity with an alignment sequence.

Chains with high identity are associated with the best-matching

sequence (however, only one chain per structure is associated with

a sequence). Multalign Viewer registers with the Chimera core’s

“model opened” trigger so that as new structures are opened, they

will also be examined and associated if appropriate. Conversely, if

the alignment sequence names are recognizable as including

SCOP

19,20

or PDB

4

identiﬁers using a few simple criteria, the

researcher can use a Multalign Viewer menu item or preference

setting to load all of the corresponding structures into Chimera.

This was easy to implement, because the Chimera core offers

functions for opening PDB/SCOP ﬁles based on their identiﬁers,

and will retrieve them via the World Wide Web or local disk as

appropriate. If Multalign Viewer fails to make an appropriate

automatic association between a sequence and structure, it can be

manually directed to make the association. Associations are indi-

cated by showing the color of the associated model behind the

name of the sequence (Fig. 3).

Once associations have been set up, many useful features

become active. Positioning the mouse over a sequence character

shows the number of the corresponding residue (in the structure) in

a status area. Making selections on the structures highlights the

corresponding regions of the sequence alignment. Dragging boxes

on the sequence alignment selects and highlights the correspond-

ing structure regions. Structure regions can be selected based on

conservation in the alignment, greatly facilitating coloring by

conservation level or showing only conserved side chains. Sec-

ondary structure elements can be depicted on the alignment with

colored boxes. Clicking on residues in the sequence makes the

Chimera window zoom in on the corresponding structure residues.

Structures can be superimposed using the sequence alignment,

optionally using only highly conserved residues, and also option-

ally, iteratively reﬁning the ﬁt by pruning poorly superimposed

residues.

An alignment can be searched with a literal string or a PROS-

ITE

21

pattern. Matches are highlighted on the alignment, and can

also be highlighted on associated structures.

Other extensions can call Multalign Viewer to show align-

ments. For example, the SSD (Structure Superposition Database)

22

extension uses Multalign Viewer to show sequence alignments

corresponding to structural alignments of interest.

Multalign Viewer is under active development. Important

short-term goals are to provide more sophisticated editing facilities

and to display and interact with phylogenetic information.

ViewDock

The ViewDock extension facilitates interactive screening of ligand

orientations from DOCK.

23,24

DOCK calculates possible binding

orientations given the structures of ligand and receptor molecules;

often, a large database of compounds is searched against a target

protein, where each compound is treated as a ligand and the target

is treated as the receptor. Simple scoring methods are used to

identify the most favorable binding modes of a given molecule and

then to rank the molecules. The output consists of a large number

of candidate ligands in the binding orientations considered most

favorable by DOCK.

Figure 3. Three structures in Chimera associated with sequences in an

alignment shown by Multalign Viewer. The sequences with color

swatches behind their names are associated with the pectate lyase struc-

tures 1jta,

45

1bn8,

46

and 2pec

47

(shown in yellow, magenta, and cyan,

respectively). The structures were superimposed using the sequence align-

ment; the ﬁts were reﬁned by iteratively removing bad residue pairings.

The sequences are colored by secondary structure (strand and helix

regions are pink and gold, respectively) and selected structure regions are

green (indicated on the structures with a green outline). Chimera’s zone

selection method was used to select all residues within 3.6 Å of the

active-site metal ion in one of the structures.

Figure 4. The ViewDock interface lists docked molecules; clicking

on a line displays just the corresponding molecule and shows its

information in the lower part of the panel. Ribose monophosphate is

shown docked to H-Ras (121p

48

). Carbon atoms are light gray, oxygen

atoms are red, nitrogen atoms are blue, and phosphorus atoms are

cyan. Hydrogens are not shown. Potential hydrogen bonds are indi-

cated with yellow lines.

1608 Pettersen et al. • Vol. 25, No. 13 • Journal of Computational Chemistry

ViewDock reads the DOCK output and provides a convenient

interface for ﬁltering results in the context of the target structure.

When a line in the list of compounds is clicked, just the corre-

sponding molecule is shown in the putative binding site, and its

information is shown in the lower part of the panel (Fig. 4). The

information may include compound name, description, and various

scores and score components. Any of the descriptors can be shown

in the list and/or used to sort it. It is also possible to view more than

one docked molecule at a time.

Compounds can be deleted from the list if visual inspection

reveals them to be unsuitable. Compounds can also be screened by

the number of hydrogen bonds formed with the target structure and

by whether or not the hydrogen bonds involve speciﬁed groups in

the site. Hydrogen-bond detection uses a set of detailed distance

and angle criteria from a published small-molecule crystal sur-

vey

25

and is an extension-provided capability of Chimera.

Movie

The Movie extension allows Chimera to show molecular dynamics

trajectories. The trajectories may be played forward or backward,

either a single frame at a time or continuously. All generic Chi-

mera capabilities such as coloring, hydrogen-bond detection,

lenses (Fig. 1), and saving PDB ﬁles are available for use with the

trajectory. Movie explicitly supports execution of a script (Python

or Chimera commands) at each frame. This makes it easy (for

example) to save images for later assembly into a QuickTime or

MPEG video.

Movie currently supports all versions of AMBER

26

trajectory

ﬁles, and support for GROMOS

27

and NAMD

28

formats is in

progress.

Volume Viewer

The Volume Viewer extension displays three-dimensional (3D)

grid data such as density maps from electron microscope recon-

structions or X-ray crystallography, calculated electrostatic poten-

tial, and solvent occupancy from molecular dynamics simulations.

It reads several ﬁle formats (CCP4

29

or MRC, BRIX or DSN6,

30

CNS

31

or XPLOR,

32

SPIDER,

33

DelPhi

34

or GRASP

35

potential

maps, PRIISM,

36

NetCDF,

37,38

and DOCK

23

scoring grids), dis

-

plays isosurfaces, meshes, and translucent solids, and allows in-

teractive adjustment of thresholds, transparency, and brightness.

Volume data is often displayed with related molecular models.

The display is automatically updated when settings are

changed. For example, dragging a threshold indicator shown on a

histogram of data values updates the displayed surface or mesh.

For large data sets, subsampling can be used to improve the

response time when the display is rotated or a threshold is changed.

Subsampling with step size 2 renders the data after omitting every

other data plane along each axis. Data sets of 256 by 256 by 256

values can be displayed with a new threshold once a second or

rotated at 10 frames per second on generic desktop PC hardware

equipped with a mid-range graphics adapter. A subregion of the

data can be chosen by dragging a box with the mouse and then

shown instead of the whole data set. Subregions can be named so

that it is easy to return to them in later sessions.

The Volume Path Tracer extension allows marker placement

and path tracing in grid data. Markers are placed by mouse-click;

the marker is positioned on the closest visible data maximum along

the line of sight under the cursor. Markers can be moved after they

are placed. Consecutively placed markers can be linked with

segments to trace a path. Additional connections can be added with

the mouse to build simple structural models. Volume Path Tracer

was developed to trace protein backbones in intermediate-resolu-

tion (5–8 Å) density maps from electron cryo-microscopy (Fig. 5)

and ﬂuorescently labeled chromosomes in 3D multiwavelength

light microscopy data (Fig. 6).

Markers and associated connecting segments are implemented

using the same mechanism as atoms and bonds. Thus, display

styles and colors can be changed in the same way as for atoms and

bonds, distances between pairs of markers and between markers

and molecular structures can be measured, and traced structures

can be aligned using Chimera’s molecule manipulation capabili-

ties. Traced paths can be displayed as smoothly interpolated curves

(Fig. 6). Markers and connecting segments can be saved in an

XML

39

ﬁle for analysis by other software.

Collaboratory

Chimera’s Collaboratory extension enables researchers at geo-

graphically distant locations to share a molecular modeling session

in real time. By default, all users connected to the same session

have equal control over the models (structures) being viewed. A

change made by any participant is immediately propagated to all

other participants, so that a synchronized view of the data is

maintained throughout a session.

Because of the complex 3D nature of molecular models, inter-

active examination of models in a real-time collaborative environ-

ment is far more effective than traditional asynchronous forms of

communication, such as passing molecular data back and forth

through e-mail. A crucial element of real-time collaboration soft-

ware is the efﬁcient transfer of information. Not only must the

software ensure that information is transmitted rapidly, it must also

consider the availability of network resources, such as bandwidth,

and transmit information in an efﬁcient format so that the appli-

cation can run in parallel with other bandwidth-heavy applications

such as videoconferencing software. Given these considerations,

application-independent collaboration tools are not as responsive

for sharing molecular modeling sessions. Such desktop-sharing

applications (e.g., Microsoft NetMeeting and Virtual Network

Computing

40

) function by transmitting the contents of the screen

from the workstation running the application to all who are sharing

it. This can be bandwidth-intensive, especially for molecular

graphics, where most operations alter a large amount of screen

content. Instead, the Collaboratory works on a lower level, by

transmitting small messages describing just the data that has been

modiﬁed.

The Collaboratory utilizes a star architecture, with a central hub

connected to multiple nodes. Each node is a user running Chimera,

while the hub is a separately running application that acts as a

rendezvous point between the nodes. Participants’ instances of

Chimera are notiﬁed when a model has been opened, closed, or

modiﬁed. A data ﬁle need only be present on the system of the

participant who opens the ﬁle in Chimera. Parameters monitored

UCSF Chimera 1609

UCSF Chimera--a visualization system for exploratory research and analysis.

Summary (2 min read)

Introduction

Chimera Architecture

Movie

Collaboratory

Results

Discussion

Acknowledgments

Figures (4)

Citations

Cites background from "UCSF Chimera--a visualization syste..."

References

Related Papers (5)