Comparing in silico platforms through the same research questions

The same research question, different computational workflows: selecting protein and small-molecule candidates

In silico research uses computers to predict molecular structures and properties and to design and evaluate candidates. Platform feature lists repeat familiar terms: structure prediction, molecular design, candidate comparison, cloud execution. Yet similar features do not mean researchers perform the same work. Some products support detailed control over calculations, some connect calculations, and some companies design and test candidates alongside research teams.

To understand these differences more concretely, we followed official guidance and public screens through two questions: “How should we choose protein candidates for experiments?” and “Which small molecules should we investigate first?” As a team building research tools ourselves, we looked closely at what each platform handles and what remains with the researcher.

Choosing protein candidates for experiments

Proteins are folded chains of amino acids. When designing a candidate to bind another protein, researchers choose a region on the target, prepare or generate candidate sequences and structures, and evaluate the results under multiple conditions.

Imagine a team that already has the target protein's structure and wants a shortlist to review before experiments. This is a hypothetical research scenario for comparison. A sequence is the order of amino acids in the protein.

From target structure preparation to candidate design, comparison of structures and conditions, and experimental priorities

“Propose new candidates” and “Make it easier to run the models we selected” call for different kinds of support.

HyperLab's Bio Co-Scientist describes a conversational approach connecting design, structure prediction, and candidate comparison. Its official antibody-design example connects backbone generation, amino-acid sequence design, and prediction of how multiple molecules are arranged together, using RFdiffusion, ProteinMPNN, and K-Fold, respectively. An antibody is a protein that recognizes and binds a particular target. In this workflow, the research team sets design conditions and reviews proposed candidates for subsequent experiments. [1]

Tamarind presents computational tools through web interfaces, batch processing, pipelines, and APIs. A pipeline passes one stage's output into the next tool. For researchers who have already chosen their methods, avoiding individual model installation can leave more time for input preparation and interpretation. The tools and settings to connect must still be chosen for the research purpose. [2]

For Neurosnap, we examined a public pipeline screen to see how tools connect. The virtual-screening template observed passes sequence and molecule lists to multiple structure-prediction tools. It illustrates reviewing prepared inputs with several tools rather than generating new protein candidates. The template-sharing guidance also states that uploaded files and protein data are not included in the code. Sharing the connections between tasks differs from sharing the actual inputs. [3]

GALUX's GaluxDesign emphasizes commissioning design and experiments together. Its public collaboration process begins with a proposed research goal, assesses feasibility, then combines design with validation in an internal laboratory. Comparing only the time spent operating software would miss the value of this collaboration. Relevant questions are who performs which research and experiments, and which candidates and data are delivered. [4]

Which small molecules should come first?

Small molecules are smaller than proteins. Given a molecule list, a team may examine possible positions and orientations at a target, properties such as solubility, and whether the molecules can be obtained or made for experiments. A target is the biological entity a drug is intended to act on.

Now imagine a team with a starting molecule and existing measurements, deciding which candidates to make or purchase next. Tools that organize data, propose candidates, or plan synthesis support different parts of that process.

From a molecule list through predicted binding poses and properties to comparison and selection for experiments

Docking searches for and scores possible poses of a molecule at a target. [5] It is an approximate method whose results depend on the target and settings. [11] Actual binding and drug effects require separate experiments. A platform should support comparison of structures, other predictions, and source data alongside the score.

Schrödinger's LiveDesign emphasizes collecting predictions and experimental data in one environment for team design and comparison. Beyond running a calculation, it supports examining new ideas against existing data. Adoption should be assessed together with the computational tools and license configuration to be connected. [6]

Standigm's STELLA Workspace describes starting with a reference molecule, generating candidates, and reviewing structures and predictions at each stage. It also describes 3D comparison and Excel export of selected candidates. The approach goes beyond a candidate list to help researchers examine what changed relative to a familiar starting molecule. [7]

Iktos describes connections between Makya for molecular design, Spaya for synthesis-route planning, and robotic synthesis and testing. Synthesis planning works backward from the desired molecule to suitable starting materials and reactions. A team wanting to make and test computational candidates should distinguish two contractual scopes: software use and commissioned experimental work. [8]

A team urgently organizing existing experimental data, a team designing new molecules, and a team without staff to operate computational infrastructure need different help. Even with the same starting data, the appropriate choice depends on which stage is being delegated.

Compare prices in equivalent units

Adoption costs are visible, but placing numbers side by side can mislead. Monthly subscriptions, computational credits, concurrency entitlements, and target-specific research services purchase different things.

HyperLab's public Core pricing, for example, applies to small-molecule lab spaces, while K-Fold and Bio Co-Scientist are listed for separate inquiries. Neurosnap distinguishes subscription plans from computational credits and describes academic and commercial usage scope. Comparing one provider's monthly fee with another's credit price cannot establish a cost difference. [9] [10]

Start with the same research task and required outcomes. Align computational workload, users, storage, export, and support scope to make pricing differences meaningful. Separately recording researcher time spent on preparation and reruns reveals burdens absent from the invoice. This time should be measured in an actual usage task rather than estimated from product descriptions.

What work remains with the research team?

Point of comparison

Records to retain consistently

What those records reveal

Input preparation

Accepted formats, conversions, errors, and corrections

Which preparation tasks the user performed

Execution

Tools, versions, settings, supported environments, waits, and failures

How environment problems affect the schedule

Interpretation

Original files, metric definitions, candidate-comparison methods

Whether evidence is sufficient to plan the next experiment

Review

Links between input and run history, export scope

Whether colleagues can understand the same result again

Costs and data

Computational units, usage scope, retention and training-use terms

Fit with budget and data-handling requirements

Features and plan-specific scope that are unclear in public materials should be discussed during adoption consultations.

Both research questions leave a common criterion: alongside model counts, examine how work is divided between your team and the platform. We aim to keep that criterion central to Nadoo Bio. By using Nadoo Cloud's computing technology to reduce the burden of managing execution environments, we want researchers to focus on questions and interpretation. The next article explains how this became Nadoo Bio's starting point.

References

  1. HITS. Bio Co-Scientist. Official product and workflow guide. Checked October 26, 2026.

  2. Tamarind Bio. Features. Official product guide. Checked October 26, 2026.

  3. Neurosnap. Pipeline Submission. Public workflow screen and usage guide. Checked October 26, 2026.

  4. GALUX. Partner with us. Official collaboration process. Checked October 26, 2026.

  5. AutoDock Vina. Basic docking. Official documentation. Checked October 26, 2026.

  6. Schrödinger. LiveDesign. Official product guide. Checked October 26, 2026.

  7. Standigm. STELLA Workspace. Official product and workflow guide. Checked October 26, 2026.

  8. Iktos. Generative AI & Robotics for Drug Discovery. Official product and research workflow guide. Checked October 26, 2026.

  9. HITS. Pricing. Official pricing. Checked October 26, 2026.

  10. Neurosnap. Products, pricing, and FAQ. Official guidance. Checked October 26, 2026.

  11. AutoDock Vina. Frequently Asked Questions. Official documentation on accuracy, approximate scoring functions, and reproducibility. Checked October 26, 2026.

Structure sources for the illustrations

The protein is based on RCSB PDB 1UBQ; the small molecule uses biotin coordinates from 1STP. These explanatory illustrations project the original coordinates with adapted colors and arrangements. Original structure publications: Vijay-Kumar et al., Structure of ubiquitin refined at 1.8 Å resolution (1987), DOI; Weber et al., Structural origins of high-affinity biotin binding to streptavidin (1989), DOI.