Experimental Design In Science

Please see this new paper I have had the pleasure of co-authoring with Dr Joanna Fox and Professor Shula Ramon entitled ‘The process of deinstitutionalisation from within an institution: evaluating innovations in a closed ward for women with (borderline) personality disorder’ (2025). This paper hits home some hard truths but it took me a long time as someone with lived experience to come to terms with accepting that facts do not care about feelings. Can inpatient services be better? Absolutely, in my decade working within in-patient I have never thought or heard ‘we are there’ or ‘we are done’ with respect to improvement. We strive to improve everyday. The idea that in-patient services do not make a difference, should be abolished and are only about containment is simply not true as demonstrated by this and many other papers. It is a disservice to staff who have dedicated their lives to supporting people and the service users to suggest otherwise, as some of the most marginalised and often discriminated in society who require a wide and dynamic range of support systems from inpatient to community. If you look at CAMHS and PD services dwindle over the years to be replaced with not even non evidence based models, just no model at all. It’s not as if we have taken away beds and made community provision better, community services have been dessimated even more with unrealistic caseloads and unprecedented levels of risk. In this paper we explore how we can deinstitutionalise from within and it’s worth having a read. 1. Purpose: • To evaluate new intervention methods alongside Dialectical Behaviour Therapy (DBT). • To explore how “deinstitutionalisation from within” can happen inside a closed ward. 2. Methods: •Photovoice (patients take photos to reflect on their experiences, then discuss them in interviews). •Staff reports every three months. •Mixed methods: qualitative (interviews, photos from service users) and quantitative (incident tracking). 3. Innovations: • Integration of Experts by Experience (people with lived experience in paid roles highly regarded by service users and staff). • Emphasis on shared decision-making and co-production. (Organisation practiced authentic Co-production at every level) • Use of peer support, “peer leave,” and activities to promote independence. • Vocational, therapeutic, and creative activities (e.g., cooking academy, mindfulness, occupational therapy clinics). 4. Findings (Interim): • Reduced incidents of self-harm and crises between evaluation periods. • Service users reported stronger self-esteem, self-worth, and empowerment. • Personalised activities and ward culture fostered trust, responsibility, and hope. • Relationships with staff described as supportive and non-judgemental, contrasting with more negative past experiences. • Transition planning (discharge books, goals, education/work ambitions) improved likelihood of successful reintegration into the community.

Could AI drafts—even imperfect ones—be a time-saver for radiologists when interpreting CT scans? Our pilot study using simulated AI reports found a 24% faster workflow, with accuracy intact. Q: What makes this study's approach unique? A: Instead of building an AI system, we used GPT-4 to simulate what AI-generated draft reports might look like. We deliberately introduced 1-3 errors in half the drafts to study how radiologists would handle imperfect AI assistance - a "Wizard of Oz" approach to prototype the future workflow. Q: How was the simulation study structured? A: We conducted a 3-reader crossover study with 20 chest CT cases. Each case was read twice: once with standard templates, and once with our simulated AI drafts. This controlled design let us directly compare the workflows. Q: What efficiency gains did you see with the simulated drafts? A: Median reporting time dropped from 573 to 435 seconds (p=0.003) - a 24% reduction. Two readers showed major improvements (717→398s and 361→322s), while one showed an increase (947→1015s). Q: Did the intentionally flawed drafts impact accuracy? A: Surprisingly, even with deliberately introduced errors in half the simulated drafts, the AI-assisted workflow showed slightly fewer clinically significant errors (0.27±0.52) compared to standard workflow (0.38±0.78). While not statistically significant, this suggests radiologists maintained their vigilance even with imperfect drafts. Q: How did radiologists respond to working with these simulated drafts? A: All 3 readers found the prototype system easy to use and well-integrated into their workflow. Two reported somewhat less mental effort, while one reported significantly reduced effort. Their likelihood to recommend it varied (scores of 5, 9, and 10 out of 10). Q: What's next? A: While these simulation results are encouraging, these are small scale pilot studies setting the stage for deeper validation. Link to short paper: https://lnkd.in/d-4aTJ69 Congratulations to stellar team of Julián Nicolás Acosta, Siddhant Dogra, Subathra Adithan, Kay Wu, MD 💫, Michael Moritz, Stephen Kwak

As academics, we all want our research to be trusted, reproducible, and strong enough to withstand review. Yet most of the problems we face during publication come from one place: weak statistical foundations and unclear experimental design. This is why I want to give you a quick, practical guide you can use to strengthen any study you are planning or refining. These principles are simple, but they prevent the most common errors I see across manuscripts, reviews, and collaborations. 1. Statistics is not about numbers. It is about reasoning. Each test, each calculation, tells a story about your data and what it truly means. 2. Experimental design begins with purpose. Define your objective clearly before you begin collecting data. The design should flow naturally from the research question. 3. Randomization protects integrity. Assign treatments randomly to eliminate bias and ensure valid comparisons. 4. Replication increases confidence. Repeating experiments strengthens conclusions and helps distinguish real effects from noise. 5. Control groups matter. They provide the baseline that gives your results meaning. Without controls, interpretation becomes speculation. 6. Choose tests based on data, not habit. Understand whether your variables are categorical, continuous, or ordinal. Then select the statistical method that fits the data, not the one that feels familiar. 7. Interpret, do not just report. Numbers are not the end of the story. Explain what they mean, why they matter, and how they support or challenge your hypothesis. 8. Visuals clarify understanding. Use tables and graphs to reveal patterns and relationships, but keep them clean, accurate, and purposeful. 9. Ethical analysis is non-negotiable. Never manipulate data to fit a narrative. Transparency and honesty sustain the credibility of your research. 10. Statistics and design are partners. Good design minimizes errors. Good statistics reveal the truth within them. One without the other cannot stand. These principles are not theoretical. They are the difference between a study that moves quickly through review and a study that struggles with rejection, uncertainty, or inconsistent conclusions. Download the full PDF below. Do you think your current research would benefit from this guide? Reply and tell me. I would love to know. ______________________________ 📌 This is Prof. Samira Hosseini. I’ve helped 12,000+ ambitious academics go from struggling with publishing papers in Q1 journals, limited visibility, and poor citation records to building a solid research trajectory and high 𝘩-index. Book a free Strategy Call, and we can dive into your challenges in top-tier journal publication and citation and see how I can best assist you: https://lnkd.in/ezqV64dX

It started as a "warm-up" project. It turned into a wake-up call for cell culture. 💡 When Cameron Movassaghi joined the lab, we devised a simple experiment plan. The goal was twofold: help him get his feet wet with proteomics, and finally answer a question that has always intrigued me: what do antibiotics actually do to mammalian cells? 🤔 We all use antibiotics (like PenStrep) in our cell cultures almost by default to prevent contamination. But are we unknowingly altering our in vitro systems? 🧫💊 Cameron took this and ran with it. He showed incredible experimental rigor, employing a longitudinal study with a crossover design. This became our lab's first deep dive with the Orbitrap Astral, allowing us to collect a massive dataset of ~9,500 proteins across 119 samples. 🔬✨ Here is what we found in our paper, just published open access in JPR: 1️⃣ Antibiotics are not silent. They rewire the proteome. We identified 108 proteins that changed abundance based on the presence of penicillin-streptomycin. 2️⃣ Metabolism Matters. The strongest hit was GCAT (glycine C-acetyltransferase), a mitochondrial enzyme involved in threonine degradation and glutathione synthesis. This suggests antibiotics drive metabolic remodeling and oxidative stress responses. ⚡ 3️⃣ Rethinking Statistics. We did a deep dive in the supplement on why these complex designs must be analyzed using Linear Mixed-Effect (LME) models rather than simple t-tests. By leveraging all samples, we proposed a model that shows how antibiotics disrupt metabolism through multiple well-known players. 📉 4️⃣ RNA does not equal Protein. Consistent with other studies, we found minimal overlap between our proteomic data and prior transcriptomic (RNA-seq) studies at the gene level. However, the story aligns clearly at the pathway level. 🧬 If you are culturing cells, this is a must-read. It turns out that "preventing contamination" might be introducing a phenotype you didn't account for. 🚫 A huge congratulations to Cameron for turning a "starter project" into a benchmark for experimental rigor in deep proteomics! 👏🎉 Paper link in the comments #Proteomics #MassSpectrometry #CellCulture #Science #Mentorship #OrbitrapAstral #Metabolism #Research

In Brief: In security, the most effective defense is often the one you don’t see –the kind that’s embedded into the very fabric of our environment. Crime Prevention Through Environmental Design (CPTED) is not just another security strategy; it is a profound shift in how we think about safety. Instead of reacting to crime, CPTED prevents it at its root –through design, visibility, and human behavior. CPTED is built on the principle that the physical environment influences human behavior –both for potential offenders and the general public. It strategically manipulates spaces to make criminal activities more difficult, more detectable, and psychologically riskier. This approach is not just theoretical; it’s been proven time and time again across urban security and corporate environments. A classic example of CPTED’s success is Bryant Park in New York City. In the 1970s and 80s, the park was a hub for drug dealing, muggings, and violent crime –a no-go zone after dark. But instead of simply increasing police patrols, the city redesigned the park using CPTED principles. ✓Natural Surveillance: Improved lighting, lower shrubbery, and open sightlines discouraged crime by eliminating hiding spots. ✓Territorial Reinforcement: The park was integrated with surrounding businesses and public spaces, giving people a sense of ownership and responsibility. ✓Access Control: New pathways and seating arrangements subtly guided movement, reducing loitering in concealed areas. The result? Crime dropped dramatically, and Bryant Park became a vibrant, safe public space –not because crime was policed out, but because it was designed out. In the corporate world, CPTED is just as powerful. A well-designed office space can significantly reduce insider threats, theft, and workplace violence. ✓Open-plan layouts and strategic workstation placements create natural surveillance, discouraging unauthorized activities. ✓Controlled access points and psychological barriers (such as clear delineations between public and private areas) reduce the likelihood of security breaches. ✓Well-lit parking lots with strategic landscaping eliminate shadows and hiding spots, reducing the risk of attacks on employees. Companies that embed security into their architecture rather than just adding cameras and guards create workplaces that feel welcoming, yet inherently secure. CPTED is not about fortification; it’s about foresight. It transforms security from reactive to proactive, from visible deterrence to invisible prevention. At a time when security threats are evolving, CPTED remains one of the most effective, cost-efficient, and sustainable crime prevention strategies available. The future of security doesn’t just rely on technology or enforcement –it relies on smarter design. #CPTED #CrimePrevention #SecurityDesign #RiskManagement #PhysicalSecurity #WorkplaceSafety #SecurityStrategy #CrimeDeterrence #PublicSafety #EnvironmentalDesign

Can you design experiments, without choosing a model up front? Classic #DesignOfExperiments assumes that it is easy to choose a good model for describing your system and that simple polynomials are up to the task (are good approximations for the true behavior). However, chemists, (cell)biologists and people who do simulations are often faced with systems where most factors might interact and show highly non-linear effects that are impossible to capture with e.g. a 2nd degree polynomial (think about e.g. pH-dependence). Using regular DOE designs in this type of system typically isn’t very fruitful. If you are working in this kind of field, you might instead opt for space-filling designs. The idea here is to cover your “field” evenly, to maximize the chance of finding optimal areas. The most common algorithm for doing this is Latin Hypercube Sampling (LHS). Lots of DOE software provides this option, but you should know that better alternatives exist, at least in some cases. LHS spreads a chosen number of runs evenly along each axis and scrambles their order. The resulting runs don’t overlap along any axis – this property is nice if some factors later turn out to have no influence and is one of the things that often make fractional factorial designs work well. But LHS has several weaknesses: 1) How do you know if 5, 10 or 25 runs is the right number? 2) The randomization often produces regions with sparse coverage. 3) LHS itself doesn’t have a good way to ensure that categoric factors (“Red”, “Green”, “Blue”) are spread evenly after randomization. An alternative that I like is “Golden Ratio Sampling” (GRS, link in comments). This algorithm is iterative (you get points one at a time), fills space very effectively and if you have categoric factors, it is easily adjusted to provide points that are space-filling for each sub-type. Pretty neat! The main weakness of GRS is that we sacrifice the lack of overlap in projection to become better at filling space. So, if you are working with 5 factors and only 2 turn out to matter, you have no guarantee of being space-filling for just these two. In fact, you always tend to get “bunching” when projecting points from GRS to lower dimensions. So which to choose? 🤷 In my opinion, when working with complex systems where domain knowledge makes you expect strong interactions and highly non-linear effects, you should go with GRS. The same goes for experiments with categoric factors. If it seems likely that most factors won’t have an effect, LHS is likely a better bet.

🚨 MIT just dropped a report saying 95% of generative AI pilots are failing. I’m not surprised one bit. I’ve lived inside the enterprise pilot world for years. The truth is: the tech isn’t the problem. The execution is. Many companies offload complex pilots to junior employees who have no experience or knowledge - that’s a recipe for failure. Here’s what we see every week: 1️⃣ Misalignment from the top down The C-suite says one thing, but the pilot team on the ground hears another. The people running the pilot almost never know what their executives truly care about. Sorry but your CFO does NOT care about you saving any of your teams time. 2️⃣ Confusing success criteria One group thinks the goal is saving time. Another group thinks it’s reducing audit/compliance risk. Another thinks it’s cost savings. Too many cooks in the kitchen → no one knows what “success” even means. 3️⃣ Data chaos Most teams don’t even know what data they need to run the pilot. By the time they scramble to collect something, it’s too late. Bad data or inaccessible data kills the pilot before it even starts. I've seen this 20x alone this year. 4️⃣ Fear So many customers go into pilots with PTSD from past failed projects. They’re already convinced it won’t work. That mindset crushes energy and momentum before anything has a chance to succeed. The bottom line: 95% failure isn’t an AI issue — it’s a LEADERSHIP and execution issue. It’s not about “trying AI.” It’s about aligning leadership, clarifying outcomes, fixing the data foundation, and shifting the mindset from fear to confidence. Because in this new world — a successful pilot requires the C-suite and the project managers to be fully aligned on what "success" is. At Terzo, we serve the largest enterprises in the world. We have learned the hard way that if the CFO is not supporting the pilot, its basically a waste of everyone's time.

Healthcare architecture is entering an incredibly transformative era. The most exciting innovations will leverage technology in a radically human-centered way to reshape how we design for care. Here are some of the most exciting spaces where I see that future emerging: Sensory-Responsive Environments AI-driven systems will adapt light, sound, and temperature in real time based on individual patient data, making recovery more personal and more effective. How might spaces that learn from each individual patient create better conditions for healing? Background Intelligence in Care Spaces Sensors and AI will handle documentation and routine tasks quietly in the background. The technology disappears and the human connection deepens. How might background intelligence change the quality of human connection in care? Mental Health-Forward Design Digital therapeutics and AI-supported behavioral tools will be embedded into environments designed around psychological safety, closing the gap between mental and physical care. What changes when technology supports emotional wellbeing as naturally as clinical monitoring? Biophilic and Data-Informed Healing Infrastructure Nature will be integrated as a clinical asset, with AI continuously informing how environments are refined for healing impact. We have always known nature supports recovery. Now we can design around it with precision. How might the natural world and intelligent systems participate in healing together? Workforce-Centered Facility Planning With AI reducing administrative burden, buildings can be designed around the people who deliver care every day, in ways that operational pressures have historically prevented. What does a facility look like when technology handles the routine and humans focus entirely on care? Pediatric and Family-Integrated Design Real-time information sharing, virtual presence, and AI-supported communication will weave families into care in meaningful ways. Children heal better when families are genuinely present. How might technology strengthen family connection rather than create distance? Decentralized Research Environments Homes and community spaces will become legitimate sites of clinical discovery through remote monitoring and AI-supported data collection, bringing research to people rather than the reverse. How might we design home environments that make people feel safe and comfortable receiving clinical care where they live? Climate-Resilient and Intelligent Infrastructure Smart building systems will help facilities manage energy, adapt to environmental stress, and serve as stable community anchors. Intelligent design makes that role permanent and intentional. How might healthcare buildings become the most trusted infrastructure in their communities?

💪🏻 𝐖𝐡𝐲 𝐒𝐞𝐪𝐮𝐞𝐧𝐭𝐢𝐚𝐥 𝐃𝐨𝐄 𝐁𝐞𝐚𝐭𝐬 "𝐁𝐢𝐠" 𝐄𝐱𝐩𝐞𝐫𝐢𝐦𝐞𝐧𝐭𝐬 ⚗️ Traditional experimental design may often follow a "big DoE" approach: plan everything upfront, run all experiments at once, then analyze. But there's a smarter way. Sequential Design of Experiments builds knowledge iteratively: • 𝐋𝐞𝐚𝐫𝐧 𝐚𝐬 𝐲𝐨𝐮 𝐠𝐨: Use early results to refine later experiments. • 𝐑𝐞𝐝𝐮𝐜𝐞𝐝 𝐫𝐞𝐬𝐨𝐮𝐫𝐜𝐞 𝐰𝐚𝐬𝐭𝐞: Stop when you have enough information, or pivot when assumptions prove wrong. • 𝐀𝐝𝐚𝐩𝐭𝐢𝐯𝐞 𝐨𝐩𝐭𝐢𝐦𝐢𝐳𝐚𝐭𝐢𝐨𝐧: Focus experimental effort where uncertainty or benefit is highest. • 𝐋𝐨𝐰𝐞𝐫 𝐫𝐢𝐬𝐤: Catch problems early rather than after completing hundreds of runs, smaller batches are easier to complete even when resources shift or priorities change. • 𝐅𝐚𝐬𝐭𝐞𝐫 𝐢𝐧𝐬𝐢𝐠𝐡𝐭𝐬: Get preliminary answers sooner, refine as needed. 🤓 In one of the use cases I have contributed to, the advantages of sequential DoE become clearly visible: at each stage, we are able to maximize the response as well as quickly reduce the variation of the response, while adapting the experimental space to new findings: modifying factors ranges, adding new factors, etc... 🤝🏻 This situation also help increasing discussions and collaboration between domain experts and design creators, where new knowledge can quickly be incorporated in the next augmentation phase. 🔄 Sequential DoE embraces uncertainty and turns it into an advantage. Why commit all resources upfront when you can learn, adapt, and optimize along the way ? ⏯️ If you are interested about this topic, I also highly recommend watching the recorded presentation 𝑩𝒊𝒈 𝑫𝑶𝑬: 𝑺𝒆𝒒𝒖𝒆𝒏𝒕𝒊𝒂𝒍 𝒂𝒏𝒅 𝑺𝒕𝒆𝒂𝒅𝒚 𝑾𝒊𝒏𝒔 𝒕𝒉𝒆 𝑹𝒂𝒄𝒆? with David Wong-Pascua, Phil Kay, Ryan Lekivetz and Ben Francis, which highlights how and why sequential DoE make the study of large experimental space (and high number of possible combinations) possible and efficient. 🔗 Link: https://lnkd.in/ercR9AQV #DoE #Learning

𝐌𝐨𝐬𝐭 𝐀𝐈 𝐩𝐢𝐥𝐨𝐭𝐬 𝐝𝐨𝐧’𝐭 𝐟𝐚𝐢𝐥 𝐛𝐞𝐜𝐚𝐮𝐬𝐞 𝐭𝐡𝐞 𝐭𝐞𝐜𝐡𝐧𝐨𝐥𝐨𝐠𝐲 𝐢𝐬 𝐰𝐞𝐚𝐤. They fail because the business is not ready to move from experiment to execution. A successful AI rollout needs more than a demo. It needs the right use case, the right environment, measurable value, production readiness, and continuous improvement. 𝐇𝐞𝐫𝐞’𝐬 𝐚 𝐬𝐢𝐦𝐩𝐥𝐞 𝟓-𝐬𝐭𝐞𝐩 𝐫𝐨𝐚𝐝𝐦𝐚𝐩 𝐞𝐧𝐭𝐞𝐫𝐩𝐫𝐢𝐬𝐞𝐬 𝐜𝐚𝐧 𝐟𝐨𝐥𝐥𝐨𝐰: 1. Choose the Right Use Case Start with a real business problem, not just an exciting AI idea. Prioritize impact, data availability, compliance, and one focused pilot. 2. Build the Pilot Environment Define the scope clearly, select the model or vendor, connect approved data sources, set guardrails, and involve real users early. 3. Validate Business Value Test real workflow scenarios, measure speed and accuracy, compare results with the manual process, and capture failures honestly. 4. Prepare for Production Assign ownership, integrate AI into existing tools, monitor quality and cost, train users, and create a rollback plan. 5. Scale and Improve Expand only after value is proven. Standardize workflows, track long-term KPIs, review model performance, and improve governance over time. AI pilots are easy to launch. Production-ready AI is harder because it requires structure, ownership, and discipline. The goal is not just to “try AI.” The goal is to make AI reliable enough for real business workflows. ♻️ Repost to help a team understand where they truly fit. ➕ Follow Prem N. more