Scientific software design: case studies on using the scientific mindset

Posted by
Gábor Lukács
on 2025-10-20

Newsletter

2025-02-20 Reading time:

Scientific software design: case studies on using the scientific mindset

 

As a chemist, you live by the scientific method. Your world is one of hypotheses, iterative optimization, and rigorous, data-driven validation. You wouldn't advance a lead compound without robust evidence, and you instinctively analyze systems to understand their underlying mechanisms.

 

Now, consider the software you use daily—from your ELN and data analysis platforms to the instrument control interfaces. Have you ever been frustrated by a clunky, illogical design? Or delighted by one that feels perfectly intuitive?

 

What you're experiencing is the outcome of User Experience (UX) design. And while it might seem like a "soft" or purely aesthetic discipline, a closer look reveals that rigorous UX is an applied science that shares a deep methodological kinship with chemistry.

 

This isn't about colors and fonts. It's about empirical research, hypothesis testing and the systematic optimization of a product to achieve a desired outcome within a complex system. That system just happens to be the human mind.

 

Let's dissect the science of UX through a chemist's lens and see how it is applied in the design of cheminformatics software.

 

Scientific software design, guide to user experience, UX in cheminformatics - Marvin, Design Hub.

 

The scientific method: a shared foundation

 

At its core, the process of developing a new drug and a new digital product follows the same iterative, scientific cycle.

Shared scientific principle Early-stage drug discovery analogue UX design analogue
Problem definition & validation Target identification & validation: Is this protein truly central to the disease pathology? Is it "druggable"? This initial validation is critical to prevent wasting millions on a flawed premise. User research & problem definition: What is the core user need or inefficiency? Is this a real, significant problem for our target user? Rigorous upfront research prevents building a solution for a non-existent problem.
Hypothesis generation High-throughput screening (HTS) & lead generation: Screening vast chemical libraries or using computational models to generate thousands of potential "hits" based on a target hypothesis. Ideation & concept development: Collaborative brainstorming, sketching and wireframing to generate a wide array of potential design solutions based on the defined user need.
Iterative refinement & optimization Lead optimization (SAR/QSAR): Systematically modifying functional groups on a lead compound and testing each analogue to optimize for potency, selectivity and ADMET properties. Prototyping & iteration: Building testable prototypes (from low-fidelity sketches to interactive mockups) and refining them based on empirical user feedback to optimize for usability, efficiency, and clarity.
Empirical validation Preclinical & clinical trials: A phased, rigorous testing process in increasingly complex biological systems (in vitro, in vivo, human) to validate safety and efficacy. Usability & A/B testing: A suite of empirical research methods to validate design effectiveness with real users in controlled or real-world scenarios.

Iterative optimization

 

The parallel in iterative optimization is particularly striking.

 

A chemist’s work in Structure-Activity Relationship (SAR) studies is the very essence of the UX iteration cycle. You synthesize an analogue, test it, analyze the data, and decide whether that modification improved your desired properties.

 

A UX designer modifies an interface element, tests it with a user, analyzes their behavior, and determines if the change reduced task time or improved comprehension. Both are methodical, data-driven optimization loops aimed at achieving a complex balance of properties.

 

Example for iterative optimization from Design Hub

 

image (13)

Initial version

 

In the initial version of the Design Hub spreadsheet, we implemented a drag & drop column reorder feature directly on the table. At the time, this solution met the existing user needs and worked well for smaller datasets.

 

However, as the application evolved and new capabilities increased the number of available columns, the horizontal drag & drop interaction became inefficient and difficult to use — users had to drag columns across a long distance. Usability testing also confirmed that this approach had poor usability when managing large numbers of columns.

image (14)Current iteration

 

To address this, we introduced a new interaction model: instead of horizontal drag & drop within the table, users can now reorder columns vertically in a dedicated side panel. This solution significantly reduces the physical movement required to reorder items and improves overall usability and control when managing large sets of data columns.

 

Designing for a "black box": biological systems vs. human cognition

 

A profound shared challenge lies in the "black box" nature of your respective target systems.

 

As a chemist, you can't directly observe every molecular interaction within a living organism. You administer a compound and observe the systemic output—cell viability, receptor binding, animal behavior—and from that data, you infer the internal mechanisms and pathways.

 

A UX designer faces the same challenge. They cannot "see" inside a user's mind. They observe external user behavior—clicks, navigation paths, task completion rates, eye movements—and must infer the internal cognitive processes: attention, memory, decision-making and frustration.

 

To navigate this, UX designers don't rely on guesswork; they rely on principles from cognitive science, which function as the "physical chemistry" of human interaction. These cognitive "rules" allow a designer to create interfaces that feel intuitive because they align with the brain's natural processing tendencies:

 

Hick's law

 

States that the time it takes to make a decision increases with the number and complexity of choices. This is why a streamlined instrument panel with 5 key parameters is superior to one presenting 50 simultaneously. It's about reducing cognitive load.

 

Miller's law

 

Suggests short-term memory is limited to roughly 7 items. This is why a multi-step synthesis protocol is best presented in a tabbed interface, breaking the process into manageable chunks rather than an overwhelming wall of text.

 

Example for Miller's law in Design Hub

image (11)

 

One of the reasons we limit the number of compounds displayed in comparison charts to a maximum of seven is Miller’s Law, which states that the average person can hold about 7 ± 2 items in their working memory. Keeping the number of compounds within this range helps users process and compare information effectively, avoiding cognitive overload.

 

Gestalt principles

 

These principles of perception (like proximity and similarity) describe how our brains inherently organize visual information. This is analogous to consistent labeling of functional groups or grouping related data points on a graph to reveal a trend.

 

The UX analytical lab: a chemist's guide to research methods


This is where UX truly reveals its scientific rigor. Forget subjective opinions; modern UX runs on empirical data gathered from a suite of analytical tools.


Usability testing as an in vitro assay


This is the workhorse of UX validation.


A user (the "system") is given a prototype (the "compound") and asked to perform a specific task in a controlled environment. Researchers observe and measure outcomes like task success rate, time on task, and error rate.


It is a direct, empirical test of your design's performance under specific conditions.


Example for usability testing from Marvin

unnamed1

 

In our case, we iterated on an existing function in Marvin after identifying a usability issue during testing. The index of the repeating unit was not visible in all cases, so we improved the function to allow the index to be repositioned more effectively.


A/B testing as a randomized controlled trial (RCT)


This is perhaps the most direct scientific analogue.


An A/B test is a controlled experiment where users are randomly assigned to two different versions of a design (e.g., 'A' is the control, 'B' is the variant with a modified button). By measuring a key metric (like conversion rate) and applying statistical analysis, designers can determine with confidence if the change produced a significant improvement.


It is hypothesis testing in its purest form.


Example for A/B testing from Marvin

unnamed2

 

We conducted an A/B test to compare two layout versions of Marvin.


Participants were divided into two groups: those who were already familiar with Marvin and those who were not. We hypothesized that both user groups would prefer the floating toolbar panel (Version B), but the results disproved this, as the other layout version (Version A) was preferred instead.


Eye-tracking as a biophysical technique


Imagine wanting to know where your fluorescently-labeled drug concentrates in a tissue sample. That's what eye-tracking does for a user interface. It provides objective, quantitative data on a user's gaze patterns, revealing what they see, what they ignore and where they struggle.


It's like having a probe that measures visual attention, allowing you to optimize a layout for maximum clarity and impact.


Analytics & heatmaps as high-content screening


Just as you might analyze a massive dataset from an HTS campaign to find patterns, UX designers use analytics tools to analyze the behavior of thousands of users. Heatmaps, which visualize where users click and scroll, are a powerful way to see aggregate behavioral trends, identifying "hotspots" of activity and "cold spots" that are being ignored.


Example for analytics & heatmaps from Marvin

unnamed3

 

We used heatmap analysis to identify which areas or functions within Marvin received the least interaction during typical chemical drawing tasks.

 

For example, the template library tool was among the least used features. Based on these insights, we decided to pack or hide such tools on the left toolbar. This is important because it helps declutter the interface, making frequently used tools easier to access.

 

The scientist's burden: navigating bias in data and design

 

Any rigorous scientific endeavor requires a constant, conscious effort to identify and mitigate bias. Your training has drilled into you the importance of controls, blinding and objective measurement to ensure the integrity of your results. UX, as an applied behavioral science, grapples with the very same cognitive biases that can distort experimental findings.

 

Here’s how these biases manifest in both worlds:

 

Bias type Impact in chemistry / scientific research Impact in UX design
Confirmation bias The tendency to seek, interpret and recall data that confirms your pet hypothesis. You might over-emphasize findings that support your proposed reaction mechanism while downplaying contradictory data from a control experiment. Seeking to validate a design rather than test it. A designer might ask leading questions in an interview, such as "Wasn't that new feature easy to use?" This pressures the user to agree and yields tainted data.
Observer bias When an outcome depends on subjective interpretation, your expectations can influence the reading. Think of unconsciously interpreting an ambiguous band on a Western blot as a positive result because you expect the protein to be there. A moderator in a usability test might unconsciously guide a user toward a "correct" action with subtle cues, or interpret a user's hesitation as confusion when it's merely thoughtfulness, skewing the observational data.
False consensus bias The tendency to assume your own choices and judgments are common. You might assume a tricky synthesis step that you've perfected is "easy" and will be simple for anyone else in the lab to replicate, failing to document critical but subtle details. A team of expert programmers designs an interface for novice users, assuming the novices will understand the same technical jargon and complex workflows that the team finds obvious. This leads to unusable products.
Negativity bias The human tendency to give more weight to negative experiences. A single failed reaction can feel more significant than ten successful ones, potentially causing you to abandon a promising research avenue prematurely due to an over-focus on a minor setback. Users remember frustrating interactions more vividly than smooth ones. This can cause designers to over-focus on fixing one loudly-reported minor annoyance while overlooking the silent success of the core features that work well for the majority.
Availability bias Relying on information that comes to mind most easily. You might opt for a synthesis method from a paper you read last week, not because it's the best, but because it's the most recent in your memory, overlooking a superior method from a more thorough literature search. Making a major design change based on a single, emotional piece of feedback from a recent support call, instead of relying on broader, more representative data from analytics or large-scale surveys.

 

The mitigation strategies are shared: rigorous experimental design, blinding where possible, asking open-ended questions, seeking disconfirming evidence and prioritizing observation of behavior over stated opinions.


It is a shared commitment to intellectual honesty.


Example for handling negativity bias from Marvin


In our clean-to-scaffold usability test, we evaluated a function that aligns multiple chemical structures based on their shared core or scaffold. The goal was to help users visualize and compare related molecules more clearly and consistently.


We used the UMUX-Lite questionnaire to understand how relevant users rated the ease of use and usefulness of this function, which helped prevent negativity bias. While participants did encounter some usability issues, they still found the clean-to-scaffold function valuable and easy to use, so we focused on improving only the most severe issues.


In case you want to do your part in shaping the future of scientific software, let us know.

 

Conclusion: a shared commitment to rigor

 

The goal of this comparison is not to diminish the unique complexities of chemistry. Rather, it is to reveal UX design as a fellow scientific discipline, one that replaces reagents and assays with prototypes and user testing, but retains the same fundamental commitment to empirical evidence and iterative improvement.

 

The chemist's mind—analytical, systematic and data-driven—is perfectly wired for this kind of thinking. By appreciating the science behind UX, you can not only become a more discerning user of your own digital tools but also apply this problem-solving framework to your own work, whether in communicating your research, designing a presentation, or collaborating on new software.

 

The pursuit of an optimal outcome within a complex system is a universal scientific challenge. Whether that system is a molecule or a mind, the principles of rigorous, evidence-based discovery remain the same.

 

Want to see the result of these methods? Check out Marvin, our chemical drawing software.

 

 

 

 

 

As a chemist, you live by the scientific method. Your world is one of hypotheses, iterative optimization, and rigorous, data-driven validation. You wouldn't advance a lead compound without robust evidence, and you instinctively analyze systems to understand their underlying mechanisms.

 

Now, consider the software you use daily—from your ELN and data analysis platforms to the instrument control interfaces. Have you ever been frustrated by a clunky, illogical design? Or delighted by one that feels perfectly intuitive?

 

What you're experiencing is the outcome of User Experience (UX) design. And while it might seem like a "soft" or purely aesthetic discipline, a closer look reveals that rigorous UX is an applied science that shares a deep methodological kinship with chemistry.

 

This isn't about colors and fonts. It's about empirical research, hypothesis testing and the systematic optimization of a product to achieve a desired outcome within a complex system. That system just happens to be the human mind.

 

Let's dissect the science of UX through a chemist's lens and see how it is applied in the design of cheminformatics software.

 

Scientific software design, guide to user experience, UX in cheminformatics - Marvin, Design Hub.

 

The scientific method: a shared foundation

 

At its core, the process of developing a new drug and a new digital product follows the same iterative, scientific cycle.

Shared scientific principle Early-stage drug discovery analogue UX design analogue
Problem definition & validation Target identification & validation: Is this protein truly central to the disease pathology? Is it "druggable"? This initial validation is critical to prevent wasting millions on a flawed premise. User research & problem definition: What is the core user need or inefficiency? Is this a real, significant problem for our target user? Rigorous upfront research prevents building a solution for a non-existent problem.
Hypothesis generation High-throughput screening (HTS) & lead generation: Screening vast chemical libraries or using computational models to generate thousands of potential "hits" based on a target hypothesis. Ideation & concept development: Collaborative brainstorming, sketching and wireframing to generate a wide array of potential design solutions based on the defined user need.
Iterative refinement & optimization Lead optimization (SAR/QSAR): Systematically modifying functional groups on a lead compound and testing each analogue to optimize for potency, selectivity and ADMET properties. Prototyping & iteration: Building testable prototypes (from low-fidelity sketches to interactive mockups) and refining them based on empirical user feedback to optimize for usability, efficiency, and clarity.
Empirical validation Preclinical & clinical trials: A phased, rigorous testing process in increasingly complex biological systems (in vitro, in vivo, human) to validate safety and efficacy. Usability & A/B testing: A suite of empirical research methods to validate design effectiveness with real users in controlled or real-world scenarios.

Iterative optimization

 

The parallel in iterative optimization is particularly striking.

 

A chemist’s work in Structure-Activity Relationship (SAR) studies is the very essence of the UX iteration cycle. You synthesize an analogue, test it, analyze the data, and decide whether that modification improved your desired properties.

 

A UX designer modifies an interface element, tests it with a user, analyzes their behavior, and determines if the change reduced task time or improved comprehension. Both are methodical, data-driven optimization loops aimed at achieving a complex balance of properties.

 

Example for iterative optimization from Design Hub

 

image (13)

Initial version

 

In the initial version of the Design Hub spreadsheet, we implemented a drag & drop column reorder feature directly on the table. At the time, this solution met the existing user needs and worked well for smaller datasets.

 

However, as the application evolved and new capabilities increased the number of available columns, the horizontal drag & drop interaction became inefficient and difficult to use — users had to drag columns across a long distance. Usability testing also confirmed that this approach had poor usability when managing large numbers of columns.

image (14)Current iteration

 

To address this, we introduced a new interaction model: instead of horizontal drag & drop within the table, users can now reorder columns vertically in a dedicated side panel. This solution significantly reduces the physical movement required to reorder items and improves overall usability and control when managing large sets of data columns.

 

Designing for a "black box": biological systems vs. human cognition

 

A profound shared challenge lies in the "black box" nature of your respective target systems.

 

As a chemist, you can't directly observe every molecular interaction within a living organism. You administer a compound and observe the systemic output—cell viability, receptor binding, animal behavior—and from that data, you infer the internal mechanisms and pathways.

 

A UX designer faces the same challenge. They cannot "see" inside a user's mind. They observe external user behavior—clicks, navigation paths, task completion rates, eye movements—and must infer the internal cognitive processes: attention, memory, decision-making and frustration.

 

To navigate this, UX designers don't rely on guesswork; they rely on principles from cognitive science, which function as the "physical chemistry" of human interaction. These cognitive "rules" allow a designer to create interfaces that feel intuitive because they align with the brain's natural processing tendencies:

 

Hick's law

 

States that the time it takes to make a decision increases with the number and complexity of choices. This is why a streamlined instrument panel with 5 key parameters is superior to one presenting 50 simultaneously. It's about reducing cognitive load.

 

Miller's law

 

Suggests short-term memory is limited to roughly 7 items. This is why a multi-step synthesis protocol is best presented in a tabbed interface, breaking the process into manageable chunks rather than an overwhelming wall of text.

 

Example for Miller's law in Design Hub

image (11)

 

One of the reasons we limit the number of compounds displayed in comparison charts to a maximum of seven is Miller’s Law, which states that the average person can hold about 7 ± 2 items in their working memory. Keeping the number of compounds within this range helps users process and compare information effectively, avoiding cognitive overload.

 

Gestalt principles

 

These principles of perception (like proximity and similarity) describe how our brains inherently organize visual information. This is analogous to consistent labeling of functional groups or grouping related data points on a graph to reveal a trend.

 

The UX analytical lab: a chemist's guide to research methods


This is where UX truly reveals its scientific rigor. Forget subjective opinions; modern UX runs on empirical data gathered from a suite of analytical tools.


Usability testing as an in vitro assay


This is the workhorse of UX validation.


A user (the "system") is given a prototype (the "compound") and asked to perform a specific task in a controlled environment. Researchers observe and measure outcomes like task success rate, time on task, and error rate.


It is a direct, empirical test of your design's performance under specific conditions.


Example for usability testing from Marvin

unnamed1

 

In our case, we iterated on an existing function in Marvin after identifying a usability issue during testing. The index of the repeating unit was not visible in all cases, so we improved the function to allow the index to be repositioned more effectively.


A/B testing as a randomized controlled trial (RCT)


This is perhaps the most direct scientific analogue.


An A/B test is a controlled experiment where users are randomly assigned to two different versions of a design (e.g., 'A' is the control, 'B' is the variant with a modified button). By measuring a key metric (like conversion rate) and applying statistical analysis, designers can determine with confidence if the change produced a significant improvement.


It is hypothesis testing in its purest form.


Example for A/B testing from Marvin

unnamed2

 

We conducted an A/B test to compare two layout versions of Marvin.


Participants were divided into two groups: those who were already familiar with Marvin and those who were not. We hypothesized that both user groups would prefer the floating toolbar panel (Version B), but the results disproved this, as the other layout version (Version A) was preferred instead.


Eye-tracking as a biophysical technique


Imagine wanting to know where your fluorescently-labeled drug concentrates in a tissue sample. That's what eye-tracking does for a user interface. It provides objective, quantitative data on a user's gaze patterns, revealing what they see, what they ignore and where they struggle.


It's like having a probe that measures visual attention, allowing you to optimize a layout for maximum clarity and impact.


Analytics & heatmaps as high-content screening


Just as you might analyze a massive dataset from an HTS campaign to find patterns, UX designers use analytics tools to analyze the behavior of thousands of users. Heatmaps, which visualize where users click and scroll, are a powerful way to see aggregate behavioral trends, identifying "hotspots" of activity and "cold spots" that are being ignored.


Example for analytics & heatmaps from Marvin

unnamed3

 

We used heatmap analysis to identify which areas or functions within Marvin received the least interaction during typical chemical drawing tasks.

 

For example, the template library tool was among the least used features. Based on these insights, we decided to pack or hide such tools on the left toolbar. This is important because it helps declutter the interface, making frequently used tools easier to access.

 

The scientist's burden: navigating bias in data and design

 

Any rigorous scientific endeavor requires a constant, conscious effort to identify and mitigate bias. Your training has drilled into you the importance of controls, blinding and objective measurement to ensure the integrity of your results. UX, as an applied behavioral science, grapples with the very same cognitive biases that can distort experimental findings.

 

Here’s how these biases manifest in both worlds:

 

Bias type Impact in chemistry / scientific research Impact in UX design
Confirmation bias The tendency to seek, interpret and recall data that confirms your pet hypothesis. You might over-emphasize findings that support your proposed reaction mechanism while downplaying contradictory data from a control experiment. Seeking to validate a design rather than test it. A designer might ask leading questions in an interview, such as "Wasn't that new feature easy to use?" This pressures the user to agree and yields tainted data.
Observer bias When an outcome depends on subjective interpretation, your expectations can influence the reading. Think of unconsciously interpreting an ambiguous band on a Western blot as a positive result because you expect the protein to be there. A moderator in a usability test might unconsciously guide a user toward a "correct" action with subtle cues, or interpret a user's hesitation as confusion when it's merely thoughtfulness, skewing the observational data.
False consensus bias The tendency to assume your own choices and judgments are common. You might assume a tricky synthesis step that you've perfected is "easy" and will be simple for anyone else in the lab to replicate, failing to document critical but subtle details. A team of expert programmers designs an interface for novice users, assuming the novices will understand the same technical jargon and complex workflows that the team finds obvious. This leads to unusable products.
Negativity bias The human tendency to give more weight to negative experiences. A single failed reaction can feel more significant than ten successful ones, potentially causing you to abandon a promising research avenue prematurely due to an over-focus on a minor setback. Users remember frustrating interactions more vividly than smooth ones. This can cause designers to over-focus on fixing one loudly-reported minor annoyance while overlooking the silent success of the core features that work well for the majority.
Availability bias Relying on information that comes to mind most easily. You might opt for a synthesis method from a paper you read last week, not because it's the best, but because it's the most recent in your memory, overlooking a superior method from a more thorough literature search. Making a major design change based on a single, emotional piece of feedback from a recent support call, instead of relying on broader, more representative data from analytics or large-scale surveys.

 

The mitigation strategies are shared: rigorous experimental design, blinding where possible, asking open-ended questions, seeking disconfirming evidence and prioritizing observation of behavior over stated opinions.


It is a shared commitment to intellectual honesty.


Example for handling negativity bias from Marvin


In our clean-to-scaffold usability test, we evaluated a function that aligns multiple chemical structures based on their shared core or scaffold. The goal was to help users visualize and compare related molecules more clearly and consistently.


We used the UMUX-Lite questionnaire to understand how relevant users rated the ease of use and usefulness of this function, which helped prevent negativity bias. While participants did encounter some usability issues, they still found the clean-to-scaffold function valuable and easy to use, so we focused on improving only the most severe issues.


In case you want to do your part in shaping the future of scientific software, let us know.

 

Conclusion: a shared commitment to rigor

 

The goal of this comparison is not to diminish the unique complexities of chemistry. Rather, it is to reveal UX design as a fellow scientific discipline, one that replaces reagents and assays with prototypes and user testing, but retains the same fundamental commitment to empirical evidence and iterative improvement.

 

The chemist's mind—analytical, systematic and data-driven—is perfectly wired for this kind of thinking. By appreciating the science behind UX, you can not only become a more discerning user of your own digital tools but also apply this problem-solving framework to your own work, whether in communicating your research, designing a presentation, or collaborating on new software.

 

The pursuit of an optimal outcome within a complex system is a universal scientific challenge. Whether that system is a molecule or a mind, the principles of rigorous, evidence-based discovery remain the same.

 

Want to see the result of these methods? Check out Marvin, our chemical drawing software.