Library › Data & Computing

Data & Computing: 190 3D models

Networks, 3D charts, AI models, servers and pipelines, sequence and omics visuals (real sizes any). Each one is a ready-made 3D model you can turn, recolour and place in a figure — open BioDiorama and type its name in the Library search.

Open the 3D Studio

3D charts · Trees & clustering · Networks & graphs · AI & machine learning · Computers & hardware · Data, files & databases · Pipelines & workflows · Sequencing & omics visuals · Lab automation & LIMS · Data privacy & security · FAIR data & sharing · Molecular simulation · Drug discovery in silico · Structure prediction & design · Metagenomics workflow

3D charts 38

3D bar chart — A grid of bars standing on a floor with grid walls — counts or values across two categories at once (for example genes × conditions).
Surface plot — A coloured 3D surface z = f(x, y) over a floor heat map — response surfaces, energy landscapes and model fits.
3D scatter plot — Points in a 3D plot box, coloured by group, with optional drop lines to the floor.
UMAP / t-SNE cell clusters — A single-cell embedding: thousands of cells drawn as dots, coloured by cluster (UMAP or t-SNE look), with small axis arrows and optional cluster numbers.
Pseudotime trajectory — Cells along a branching differentiation path, coloured by pseudotime, with the fitted principal curve and velocity-style arrows showing direction.
Volcano plot — Differential expression: fold change against significance, with up- and down-regulated genes coloured and dashed thresholds.
Manhattan plot — Genome-wide association results: −log₁₀ P for variants along chromosomes 1–22 and X in alternating colours, with towers at associated loci and the genome-wide significance line.
Kaplan–Meier survival curve — Survival steps over time for two to four groups, with tick marks for censored cases and optional shaded confidence bands.
ROC curve — Receiver operating characteristic: sensitivity against 1 − specificity for one to three classifiers, the chance diagonal and an optional shaded area under the curve.
PCA plot (3D) — Samples in principal-component space: three labelled axes through the origin, coloured groups with translucent confidence ellipsoids and optional loading arrows (biplot).
Clustered heat map — A grid of tiles coloured by value (for example gene expression across samples), clustered into blocks, with dendrograms on two sides; tiles can rise with their value.
Correlation matrix — The lower triangle of a correlation matrix: a disc for each pair of variables, sized and coloured by the correlation coefficient (blue negative, red positive).
Violin plot — Rounded violins showing the distribution of values in each group, with a box for the middle half, the median dot and optional data points.
Box-and-whisker plot — Solid boxes for the quartiles of each group with median band, whiskers, outliers and the individual data points jittered alongside.
Ridgeline plot — Stacked density curves, one ridge per group or time point, set one behind another like a mountain range.
3D donut / pie chart — A solid ring or pie split into proportions, with an optional pulled-out slice and slices that rise with their share.
Contour map (terraced) — A landscape cut into flat terraces at contour levels, like a topographic model — for density or energy landscapes seen as levels.
Gene expression dot plot — Marker genes against cell groups: dot size shows the share of cells expressing each gene and colour its average level.
Waterfall plot (tumour response) — Bars for each patient’s best change from baseline, sorted from growth to shrinkage and coloured by response category, with ±20 % / −30 % lines.
Forest plot — Meta-analysis: each study’s effect size as a square sized by its weight with a confidence-interval line, the pooled diamond and the line of no effect.
Sankey / alluvial diagram — Flows between three stages drawn as solid ribbons whose width shows the amount, coloured by where they start — cell-type transitions, patient pathways, read assignments.
Chord diagram — Groups as arcs around a ring with ribbons across the middle showing how much each pair shares or exchanges (interactions, migrations, cell–cell signals).
Radar (spider) chart — Values on several axes joined into a polygon for each series, over a web of rings — profiles of samples, models or patients across measures.
Venn diagram (3D) — Two or three overlapping sets as translucent discs or spheres — shared genes, overlapping hits or common samples.
UpSet plot — Set intersections for many sets: bars for each intersection size above a dot matrix showing which sets take part, with set sizes on the side.
Scatter plot with regression line — Paired measurements with the least-squares fit line and its shaded 95 % confidence band; residual lines optional.
Dot plot with mean ± SD (SuperPlot) — Every measurement as a dot, coloured by biological replicate, with the replicate means as large dots and the mean ± SD bar for each condition.
Polar area (rose) chart — Wedges around a circle whose length and height grow with their value — seasonal, circadian or directional data.
Treemap (3D) — Nested blocks whose area shows each item’s share within its group, raised a little by value — composition of taxa, gene categories or budgets.
Parallel coordinates plot — Each sample as a line crossing several vertical axes, coloured by group — patterns across many variables at once.
Pair plot (scatter-plot matrix) — A grid of small scatter plots for every pair of variables, with density curves on the diagonal, coloured by group.
Histogram with density curve — Touching bars counting values in bins, with a smooth density curve on top; an optional second group sits behind.
Ternary plot — A triangle for three-part compositions that add up to 100 % (for example cell states or allele fractions), with grid lines and coloured points.
Vector field (3D arrows) — Arrows on a 3D lattice showing direction and strength of a flow or gradient — velocities, forces or RNA-velocity-style fields — coloured by magnitude.
3D bubble chart — Large bubbles on stems above a floor grid: position for two variables, height for a third, bubble size for a fourth and colour for group.
Waffle chart — A 10 × 10 grid of blocks, one per percent, coloured by category — shares that are easy to count.
3D density (nested shells) — A cloud of points wrapped in nested translucent shells at three density levels, with an opaque core — 3D kernel density or confidence regions.
Slope chart (before / after) — Paired values for each subject joined between two axes, coloured by whether they rise or fall — paired samples, treatment effects.

Trees & clustering 6

Phylogenetic tree (rooted) — A rooted rectangular tree: right-angled branches whose length shows change, tips coloured by clade, with an optional scale bar and clade boxes.
Phylogenetic tree (circular) — A circular (fan) tree with branches radiating from the root, tips coloured by clade and an optional outer ring of trait blocks.
Unrooted tree (radial) — An unrooted tree drawn with the equal-angle layout: straight branches spreading like a starburst, with clades in colour — common for viral and bacterial strains.
Dendrogram (hierarchical clustering) — A clustering tree of U-shaped links joining samples by similarity, leaves coloured by cluster and a dashed line where the tree is cut.
Tanglegram (co-phylogeny) — Two trees facing each other with lines linking matching tips — host and parasite, gene and species trees, or two clusterings compared.
Haplotype network — Haplotypes as discs sized by how often they occur and split by population, joined by lines with a tick for each mutation step.

Networks & graphs 12

Protein–protein interaction network — An interaction network in 3D: proteins as balls sized by how many partners they have, hubs standing out, coloured by module or by degree.
Gene regulatory network — Transcription factors (diamonds) switching target genes (balls) on or off: green arrows for activation, red bars for repression.
Signalling pathway graph — A pathway drawn as a graph from ligand and receptor at the membrane through adaptors and a kinase cascade to transcription factors and target genes in the nucleus.
Biomedical knowledge graph — Different kinds of entity — genes, drugs, diseases, pathways and phenotypes — drawn as different shapes and linked by relations in a 3D graph.
Network with communities — A network whose nodes fall into tightly linked communities (modules), each in its own colour inside a soft halo, with a few bridges between them.
Bipartite network — Two kinds of node in two rows — for example drugs and their protein targets, or microbes and metabolites — with links only between the rows.
Circular network layout — Nodes evenly spaced on a ring and links drawn as curves bowing through the middle, coloured by the module they start from — easy to compare who connects to whom.
Hive plot — Nodes placed on three axes by type and ranked by degree, with curved links between axes — a network drawn reproducibly, without a hairball.
Cell–cell communication network — Cell types as large nodes around a circle with curved arrows whose thickness shows signalling strength (ligand–receptor interactions), including autocrine loops.
Multilayer (multiplex) network — The same entities in stacked layers — for example genes in a co-expression, an interaction and a regulatory layer — with dashed links joining each node across layers.
Arc diagram — Nodes in a row with links drawn as arcs above them, coloured by group — compact for ordered things such as genes along a chromosome or residues in a sequence.
Global collaboration network — A globe of latitude and longitude lines with sites on its surface joined by raised great-circle arcs — data sharing, consortia or sample flows between centres.

AI & machine learning 30

Neural network (layers of nodes) — A fully connected neural network: input, hidden and output layers of nodes joined by links, flat or with each layer as a 3D grid; dropout can grey out some nodes.
Convolutional neural network (CNN) — An image passing through convolution and pooling blocks that shrink in size and grow in depth, then dense layers to the class output — with kernel projection lines.
Convolution (kernel sliding over an image) — A 3 × 3 kernel window on an input grid, its weights shown beside it, projecting down to one cell of the smaller output feature map.
Transformer model (stacked blocks) — A transformer: token embeddings enter a stack of blocks, each with multi-head attention, add-and-norm and a feed-forward layer with residual paths, ending in output probabilities.
Attention between tokens — Two rows of tokens with lines from each query to the keys it attends to; line thickness and colour show attention weight.
Generative adversarial network (GAN) — A generator turns random noise into a fake image; a discriminator compares it with real images and outputs real or fake — the two trained against each other.
Autoencoder / VAE — An encoder squeezes an image down to a small latent code and a decoder rebuilds it; the variational version splits the code into mean and spread and samples from it.
Latent space (manifold) — A curved sheet standing for a model’s learned latent space, with clusters of encoded samples on it and an interpolation path walking from one cluster to another.
Diffusion model (denoising steps) — A row of pictures going from pure noise to a clean image step by step, as a diffusion model removes noise; arrows show the reverse (denoising) direction.
Recurrent network (RNN / LSTM) unrolled — A recurrent cell with its loop, and the same cell unrolled over time steps: inputs below, hidden states passed along, outputs above; the LSTM version adds the cell-state line and gates.
Graph neural network (message passing) — A small graph whose nodes carry feature vectors; messages travel along the edges into one node, which combines them to update its own features.
Decision tree (classifier) — A tree of yes/no questions: split nodes ask about one feature, branches lead to leaves that give the predicted class.
Random forest (ensemble vote) — Several small decision trees each give a prediction, and the forest takes the majority vote — an ensemble model.
Support vector machine (separating plane) — Two classes of points in 3D split by a separating plane, with the margin planes either side and the support vectors ringed.
k-means clustering — Points grouped around k centroids (crosses), each point joined to its centroid by a thin spoke and coloured by its cluster.
Gradient descent on a loss landscape — A wire-frame loss surface with the path an optimiser takes downhill, step by step, to the minimum.
Training and validation loss curves — Loss falling over training epochs for the training and validation sets; where validation loss turns back up the model starts to overfit, marked by an early-stopping line.
Confusion matrix — Predicted against true classes as raised tiles with counts printed on them — the diagonal holds the correct predictions.
Artificial neuron (perceptron) — One artificial neuron: inputs multiplied by weights (line thickness), summed (Σ), passed through an activation function and sent out as the output.
Token embeddings (vectors) — Tokens (words, k-mers or genes) each turned into a column of numbers — an embedding vector — shown as coloured cells.
Reinforcement learning loop — An agent acts on an environment (a grid world with walls and a goal) and gets back the new state and a reward — the loop it learns from.
Federated learning — Hospitals or labs each train a model on their own data, which never leaves the site (locked); only model updates travel to a central server, which sends back the combined model.
Tensor (3D array) — A three-way array of numbers as a cube of cells — samples × genes × time, or an image batch — with one slice pulled out.
AI accelerator chip — An AI accelerator package: a large processor die covered in a grid of compute cores, flanked by stacked high-bandwidth memory, on an interposer and substrate. 1 unit = 1 cm. (≈ 5.5 cm package)
GPU compute node (lid off) — An open 4U server packed with eight GPU modules under finned heat sinks, a wall of fans at the front, two CPUs with memory and power supplies at the back. (≈ 44 × 80 × 18 cm)
U-Net (image segmentation network) — The U-shaped network for segmenting images: feature blocks shrink down the left side, grow back up the right, with skip connections across, from a microscopy image to its mask.
Vision transformer (image patches to tokens) — An image cut into a grid of patches, each patch flattened into a token with its position number and fed, with a class token, into a transformer encoder.
Activation function plot — The curve of a neural-network activation function — ReLU, sigmoid, tanh, GELU or softplus — on small axes; several can be overlaid.
Hyperparameter grid search — A 3D grid of training runs over three settings (learning rate, depth, batch size), each ball coloured by its score, with the best run ringed.
AI cell segmentation (masks) — A microscopy field with every cell found by a segmentation model, each outlined or filled with its own colour mask.

Computers & hardware 17

Rack server (1U, lid off) — A one-unit rack server with its lid removed: drive bays at the front, a row of fans, two processors under heat sinks with memory sticks either side, and power supplies at the back. (≈ 44 × 70 × 4.4 cm)
HPC cluster (row of racks) — A row of open server racks filled with compute nodes and blinking lights, linked by an overhead cable tray — a high-performance computing cluster or data-centre aisle. (≈ 60 cm wide, 2 m tall racks (1 unit ≈ 1 m))
Laptop with data analysis — An open laptop with an analysis on screen — clusters, code beside a plot, a genome browser, a heat map or a network (setting). (≈ 32 cm wide)
Desktop workstation (two screens) — A tower computer beside two monitors on stands, one with code and one with results, plus keyboard and mouse — a dry-lab or analysis workstation. (≈ 1.2 m across)
Cloud with data flowing — A cloud receiving and sending data packets along curved paths to a laptop, a sequencer and a server below — cloud computing and storage.
Hard disk drive (opened) — A 3.5-inch hard drive with its cover off: mirror-bright platter on the spindle, the actuator arm with its read/write head, the magnet and the case. 1 unit = 1 cm. (10.2 × 14.7 × 2.6 cm)
Solid-state drive (M.2) — An M.2 NVMe solid-state drive: a slim board with flash-memory chips, a controller and cache, the gold edge connector with its notch and the screw slot. 1 unit = 1 cm. (2.2 × 4.2–8.0 cm)
Data tape cartridges — Magnetic tape cartridges for long-term archiving, one with its shell shown see-through to reveal the reel of tape. 1 unit = 1 cm. (10.2 × 10.5 × 2.15 cm each)
Terminal window (command line) — A command-line terminal floating as a panel, with real commands on screen: read alignment, a cluster job, a workflow run or version control.
Code editor window — A floating code-editor panel with tabs, line numbers, syntax-coloured lines of code (drawn as coloured bars) and a minimap.
Computational notebook — A notebook panel with a heading, code cells and their results — a bar chart and a scatter plot — like a Jupyter or R Markdown notebook.
Quantum computer (dilution refrigerator) — The gold “chandelier” of a superconducting quantum computer: stacked cooling plates joined by rods, coiled coaxial lines running down to the processor at the bottom, with an optional outer shield. (≈ 1.5 m tall chandelier (1 unit ≈ 30 cm))
Network switch with cables — A rack-mount network switch with two rows of ports and status lights, patch cables in several colours plugged in and looping away. (≈ 44 cm wide)
Fibre-optic bundle (data light) — A cable splitting into a fan of glass fibres with glowing tips — high-speed data links between machines and sites.
Visualisation wall (tiled screens) — A wall of tiled screens with thin bezels showing different results at once — clusters, a genome browser, heat maps and networks — for a data-visualisation lab or control room. (≈ 2 m wide per 3 screens (1 unit ≈ 1 m))
Cloud storage bucket — A metal bucket full of data blocks and files, under a small cloud — object storage for raw data, images and results.
Network storage array (NAS) — A storage server with a front of hot-swap drive trays and status lights, one tray pulled half out — shared lab storage. (≈ 30 cm wide)

Data, files & databases 9

Database (stacked drums) — The classic database symbol made solid: stacked drums with rims, alone or as a replicated cluster linked together.
Data table (spreadsheet) — A floating table of tiles: a coloured header row, an index column and cells holding value bars, with one row highlighted — tidy data, a sample sheet or a count matrix.
Data file (FASTQ, BAM, VCF, CSV…) — A document with a folded corner, text lines and a coloured tag naming its format — raw reads, alignments, variant calls, tables, images and more; a stack can sit behind it.
Folder of files — An open folder with sheets and a data file sticking out — a project directory, dataset or shared drive.
Binary data stream — A ribbon of tiles, each carrying a 1 or a 0, curving forward and growing as it comes — digital data, a byte stream or data transfer.
Data lake — A lake basin filled by streams of incoming data, with raw files and blocks of every kind floating in it — one store for raw, unstructured data.
Database schema (linked tables) — Three or four tables, each with a coloured header and rows of fields, joined by relationship lines with one-to-many (crow’s foot) ends — a relational data model.
Compressed archive (zip) — A file with a zipper down its middle and a pull tab — a compressed archive (zip, tar.gz) for sharing or storing data.
Data backup and sync — Two data stores with curved arrows chasing each other between them — backup, replication or synchronisation.

Pipelines & workflows 10

Workflow DAG (pipeline steps) — Pipeline steps as blocks joined by arrows in a directed acyclic graph, coloured by status — done, running, waiting or failed — as in Snakemake, Nextflow or Airflow.
Software containers (stacked) — Shipping containers stacked like a container ship’s load — software containers (Docker, Apptainer) that package tools so analyses run the same anywhere.
Container orchestration cluster — Hexagonal worker nodes each running a few containers (pods), managed by a central control node — Kubernetes-style scaling of analysis jobs.
Data pipeline (sources to warehouse) — Three data sources feed clear pipes carrying data blocks into a processing unit with gears, which fills a large data store — extract, transform, load.
Version control (branch graph) — A history of commits as dots on lanes, with a feature branch splitting off and merging back, and a tagged release — version control with git.
Quality-control filter funnel — Many data points poured into a funnel; fewer, clean ones come out below while failed ones (red) drop into a reject tray — filtering and QC.
Flowchart (decision steps) — A flowchart with start and end capsules, process boxes, a decision diamond and yes/no arrows — analysis plans, protocols and algorithms.
API connection (plug and socket) — A plug on a cable about to click into a socket block — an API or interface connecting two tools or services.
Job queue to a cluster — Jobs waiting as coloured blocks on a conveyor that feeds a small cluster of servers — batch scheduling on an HPC system (Slurm, PBS).
Parallel processing (split, map, reduce) — A big dataset split into chunks, each processed in parallel by its own worker, the partial results shuffled and combined into one answer — scatter-gather or map-reduce.

Sequencing & omics visuals 25

Sequencing reads aligned to a genome — Short reads stacked under a reference sequence, as in a genome viewer: forward and reverse reads in two colours, a variant column where many reads disagree with the reference, and coverage above.
Genome browser tracks — Stacked genome tracks over one region: a ruler, a gene model with exons and introns, RNA-seq coverage for two samples, ChIP-seq peaks and variant lollipops.
k-mers from a sequence — A DNA sequence as letter tiles with its overlapping k-mers lifted out above it one step at a time — the words used for counting, assembly and alignment-free methods.
Genome assembly (reads to contigs) — Overlapping reads, coloured by where they belong, merging into longer contigs below, joined into a scaffold with gaps.
De Bruijn assembly graph — k-mers as nodes linked where they overlap by k − 1 letters, with a bubble from a sequencing error or variant and a loop from a repeat.
Sequence logo — Stacked letters at each position of a motif, each letter as tall as its frequency times the information content — a transcription-factor binding site or splice site.
Multi-omics layer stack — Genome, transcriptome, proteome and metabolome as stacked translucent layers, each with its own molecules as nodes, joined by lines between layers — multi-omics integration.
Circos genome plot — Chromosomes as arcs around a ring, inner tracks of data bars and points, and ribbons across the middle for rearrangements or interactions between chromosomes.
Multiple sequence alignment — Aligned protein or DNA sequences as rows of coloured letter tiles with gaps, and a conservation bar chart above each column.
Sanger sequencing chromatogram — Four coloured fluorescence traces with a peak for each base and the base calls printed above — a capillary (Sanger) sequencing read.
Genome ideogram with hits — All chromosomes as banded bars side by side (an ideogram), with coloured markers beside them for hits — GWAS loci, variants or gene locations.
Hi-C contact map (TADs) — A triangular Hi-C contact map: diamond tiles coloured by how often two parts of a chromosome touch, with domains (TADs) outlined along the diagonal.
Spatial transcriptomics spots — A tissue section covered by a hexagonal array of capture spots, each coloured by its gene-expression cluster — spatial gene expression on a slide.
Sequencing read structure — The parts of a library read laid end to end: adapters, cell barcode, UMI, poly(T) and the cDNA insert (single-cell) or index and insert (paired-end Illumina).
Mass spectrum (peaks) — A stick mass spectrum: peaks along m/z with their heights as intensity, the tallest labelled — proteomics or metabolomics data.
Synteny plot — Two chromosomes from different genomes, one above the other, with ribbons linking matching blocks — one block flipped to show an inversion.
Gene set enrichment (GSEA) plot — The running enrichment score along a ranked gene list, a barcode of where the set’s genes fall, and the ranking metric shaded from red to blue below.
Mutation lollipop plot — A protein drawn as a bar with its domains, and lollipops at mutated positions: stem height for how often, head colour for the kind of change (missense, truncating, in-frame).
OncoPrint (mutation matrix) — Genes against tumour samples as a grid: grey cells with coloured marks for amplification, deletion, missense and truncating changes, sorted so altered samples sit together, with a bar of alterations per sample on top.
Pangenome graph — Several genomes threaded as coloured paths through a shared graph of sequence blocks, splitting into bubbles where they differ — a pangenome or variation graph.
DNA methylation reads (lollipops) — Single reads as lines with a lollipop at each CpG site — filled when methylated, open when not — as from bisulfite or long-read sequencing.
Sashimi plot (splice junctions) — RNA-seq coverage over a gene’s exons with arcs joining the exons that reads splice between, each arc as thick as its read count and labelled — here one exon is often skipped.
Nanopore signal (squiggle) — The raw electrical current from a nanopore read: a noisy step trace where each level reflects the k-mer sitting in the pore, with the called bases beneath.
Taxonomy sunburst (Krona-style) — Rings of nested wedges from phylum in the middle to genus at the edge, each wedge sized by abundance and stepping up in height — a metagenome’s composition.
Microbiome composition (stacked bars) — One bar per sample split into the relative abundance of each taxon, adding up to 100 % — gut or soil microbiome profiles compared.

Lab automation & LIMS 6

Lab robot arm moving a microplate — A compact collaborative robot arm on a bench plate, its gripper carrying a 96-well plate between two plate nests — the hands of an automated lab. (≈ 1 m reach (1 unit = 10 cm))
Automated lab workcell — An enclosed automation cell: a robot arm on a rail serving a plate hotel, an incubator, a plate reader and a dispenser, behind clear panels — a self-driving lab. (≈ 2 × 1 m cell (1 unit = 30 cm))
LIMS station (sample tracking) — A monitor showing a laboratory information system — sample list with status chips — with a hand-held barcode scanner reading a tube and a label printer printing the next label. (≈ 80 cm across (1 unit = 10 cm))
2D-barcode rack scanner — A flat-bed reader under a rack of 96 barcoded storage tubes, with its screen showing every position read — sample inventory for a biobank or freezer. (≈ 18 × 13 cm (1 unit = 10 cm))
Plate hotel (stacker) — A tall tower of slots holding microplates for a robot to fetch, one plate half pulled out, with a status light — plate storage in an automated line. (≈ 15 × 10 × 30 cm (1 unit = 10 cm))
Mobile lab robot — A wheeled robot with a laser scanner on top, a light strip and a small arm, carrying microplates between instruments — lab logistics and self-driving labs. (≈ 70 cm long (1 unit = 30 cm))

Data privacy & security 11

Locked data (padlock on a database) — A database stack with a large padlock in front, shut or open — protected, access-controlled or encrypted data.
Shield protecting data — A raised shield with a check mark standing in front of a stack of records — data protection, privacy by design and compliance.
Encrypted data block — A cube of scrambled data blocks with a key turning in its lock — encryption at rest, ciphertext and keys.
Encrypted USB drive (keypad) — A hardware-encrypted USB drive with a numeric keypad and lock light, its cap pulled off — moving sensitive data safely. 1 unit = 1 cm. (≈ 2 × 7 cm)
De-identification of records — Patient or participant records with names and faces on the left, and the same records on the right with identities masked and replaced by coded IDs.
Firewall (filtering traffic) — A brick wall with a gate: green data packets pass through, red ones are stopped and bounce back — network security and filtering.
Data vault (safe) — A heavy safe with its round door swung open, a combination dial and handle, and a database stored inside — secure storage for the most sensitive data.
Access card and reader — An ID badge with photo and chip on its clip, held up to a wall-mounted card reader showing a green light — controlled access to labs, servers and data rooms.
Differential privacy (added noise) — Summary bars with their exact tops hidden under a cloud of random noise — published statistics that cannot reveal any one person.
Computing on encrypted data — A locked glass box with gears turning inside: locked data goes in, locked results come out, and nobody sees the raw values — homomorphic encryption or secure enclaves.
Signed consent form — A consent form on a clipboard with ticked boxes for each use of the data and a signature line, a pen beside it — informed consent for data and sample use.

FAIR principles blocks — Four letter blocks spelling FAIR — Findable, Accessible, Interoperable, Reusable — for data stewardship and open science.
Data repository (archive) — A classical archive building with columns and a pediment, data drums on its steps — a public repository or data archive (GEO, SRA, Zenodo, Dryad).
Open data (box opened) — A box with its lid flipped open and data blocks rising out of it, an unlocked padlock beside — data released openly for anyone to reuse.
Dataset with DOI tag — A stack of data blocks with a luggage-style tag tied on, printed “DOI” — a citable dataset with a persistent identifier.
Interoperable data (puzzle pieces) — Four jigsaw pieces in different colours fitting together, one lifted into place — data and tools that work together through shared standards.
Data management plan (checklist) — A clipboard checklist with ticked items and a small database badge — planning how data will be stored, described, shared and kept.
Data commons hub — A shared data store in the middle linked by two-way paths to institutions around it, each with its own building — a data commons or consortium platform.
Research data lifecycle — A ring of curved arrows around a dataset — plan, collect, process, analyse, share and reuse — the life of research data.
Finding a dataset (magnifier) — A magnifying glass over a grid of dataset cards, enlarging the one that matches — searchable catalogues and rich metadata make data findable.
Ontology (term hierarchy) — Terms as labelled boxes in a hierarchy, each child linked to its parent by an “is a” arrow — controlled vocabularies such as cell, tissue or disease ontologies.

Molecular simulation 8

Membrane protein in a simulation box (molecular dynamics) — The β2-adrenergic receptor (PDB 2RH1, with carazolol and its crystal cholesterols) set in a POPC bilayer with water and Na⁺/Cl⁻ in a periodic box, as built for an all-atom or Martini coarse-grained run. The front half of the box is cut away to show the receptor in the membrane. (box ≈ 8 × 8 × 9 nm)
Solvated protein in a periodic box — Hen egg-white lysozyme (PDB 1AKI) solvated in explicit water inside a cubic, rhombic-dodecahedral or truncated-octahedral periodic cell, with eight chloride ions to neutralise its +8 charge. A wedge is cut away to show the protein and the first hydration shell around the protein. (box ≈ 6–7 nm across)
Coarse-grained mapping of a protein (atoms to beads) — Ubiquitin (PDB 1UBQ) half in atoms and half in coarse-grained beads: in Martini about four heavy atoms become one bead (one backbone bead per residue, rings split into three), while the simplest models keep two beads per residue, backbone and side chain. Bead colours follow polar, apolar, charged and ring classes. (ubiquitin ≈ 3 nm)
Protein folding energy landscape (folding funnel) — A rugged funnel of free energy: unfolded chains of the villin headpiece subdomain HP35 wander down from the wide, high-entropy rim through a partly folded intermediate to the native three-helix fold at the bottom (PDB 1YRF), the fast folder that long simulations fold from scratch. The surface is schematic. (schematic; HP35 ≈ 1.5 nm, drawn enlarged)
Elastic network model with its slowest normal modes — T4 lysozyme (PDB 2LZM) as one bead per residue joined by springs to every neighbour within reach — an anisotropic network model. Arrows show one of its three slowest normal modes, computed from this network (13 Å cut-off); beads are coloured from rigid (blue) to mobile (red), and mode 1 is the hinge bending that closes the active-site cleft. (T4 lysozyme ≈ 5 nm)
Markov state model of a conformational change — Adenylate kinase in three states joined by transition arrows, as a Markov state model summarises a long simulation: open (PDB 4AKE), closed on the inhibitor Ap5A (PDB 1AKE), and a halfway state between them. The LID (red) and NMP (gold) domains close over the CORE (blue); the middle state is an interpolation, not a deposited structure. (adenylate kinase ≈ 5 nm each)
QM/MM model of an enzyme active site — HIV-1 protease with indinavir (PDB 1HSG) split for a hybrid calculation: the catalytic Asp25 and Asp25′ side chains and the inhibitor's central hydroxyl form the quantum (QM) region inside the blue envelope, capped by hydrogen link atoms where bonds are cut; the rest of the protein is treated by molecular mechanics, its nearby charged atoms polarising the QM region. (HIV-1 protease ≈ 5 nm)
Potassium ions crossing a channel filter (simulation) — Two opposite subunits of the KcsA channel (PDB 1BL8) in its membrane, with the TVGYG selectivity filter as atoms and K⁺ in its binding sites. Simulations disagree on how ions move: alternating with water (water-mediated knock-on) or touching one another (direct knock-on); the gold trace follows one ion from the cavity out through the filter. (KcsA ≈ 5 nm tall)

Drug discovery in silico 3

Drug docked in its binding pocket — A drug in the real binding pocket of its target, as a docking program places it: a wedge of the protein is cut away so the pocket shows in section, its surface coloured by polar and non-polar atoms, with the contacting residues as sticks and the hydrogen bonds dashed. Targets are crystal structures (Abl–imatinib, HIV-1 protease–indinavir, acetylcholinesterase–donepezil). (pocket ≈ 3 nm across)
Pharmacophore model of a bound drug — The features a drug uses to bind, read from its crystal pose: hydrogen-bond donors (green) and acceptors (red) pointing at their partners, hydrophobic spots (yellow), aromatic rings (purple), a positive ionisable amine (blue), and grey exclusion volumes where the pocket's atoms leave no room. Such a model is the query for a pharmacophore screen. (ligand ≈ 1.5–2 nm)
Virtual screening funnel ending in a docked hit — A compound library narrowed stage by stage — property filters, docking, rescoring and simulation — until one hit sits docked in the target pocket (here imatinib in Abl kinase, PDB 1IEP). The molecules are real drug structures drawn to scale with one another; the funnel itself is schematic. (schematic, molecules ≈ 1 nm)

Structure prediction & design 4

Protein language model folding a sequence — The 76 residues of ubiquitin as tokens around a ring, the attention links that a protein language model learns between residues in contact, and the folded chain the model outputs (the crystal fold, PDB 1UBQ). Each token is tied to its residue in the structure, coloured from N- to C-terminus. (schematic around ubiquitin (≈ 3 nm))
Coevolving residue pairs bridging a fold — H-Ras (PDB 5P21, with GppNHp and Mg²⁺) with struts joining residues far apart in sequence but touching in space — the pairs that coevolve across homologues and that direct-coupling analysis recovers from a sequence alignment, which made contact-guided structure prediction possible. (Ras ≈ 4 nm)
Atomic model fitted into a cryo-EM density map — Human apoferritin, the 24-subunit test specimen of single-particle cryo-EM (PDB 6Z6U, 1.25 Å), shown as a density map with a wedge removed so the fitted chains show inside it. Raising the contour level thins the map, as when fitting a model into a reconstruction. (apoferritin ≈ 12 nm)
De novo designed miniprotein binder on its target — LCB1, a 56-residue three-helix miniprotein designed by computer to block SARS-CoV-2, bound to the spike receptor-binding domain at the ACE2 site (cryo-EM, PDB 7JZU). The Design step setting scatters the binder's backbone as a diffusion model starts from noise, the way newer methods generate such binders. (RBD ≈ 5 nm, binder ≈ 2.5 nm)

Metagenomics workflow 1

Metagenome-assembled genomes binned from contigs — The end of a shotgun metagenome: reads are assembled into contigs, and contigs that share coverage across samples and sequence composition are binned into draft genomes (MAGs), one per species. Each ring is one MAG drawn as a circular chromosome made of its contigs — gaps between them are the unassembled parts, and the missing stretch is what completeness estimates leave out. (genomes of 2–6 Mb)