Akshatha Nayak

It's amazing to witness how the field has evolved due to AI asssisted coding. What took us years, is now doable in a day or less! Dhirendra Kumar.

16

In 2010, building a standalone MS/MS search tool took nearly 2 years of work by two graduate students, Amit Yadav and me. Published in the Journal of Proteome Research: https://lnkd.in/ez2U22zU Recently, AI helped me create a standalone MS/MS search tool in about 2 hours using structured prompting: https://lnkd.in/ejYU3bFE The speed shift is undeniable. AI removes much of the programming barrier. But it does not replace: • Statistical rigor • Proper benchmarking • Positive and negative controls • Deep understanding of MS/MS fragmentation • Careful FDR evaluation As the job market shifts rapidly, the best path forward is not resisting AI. It is strengthening our domain knowledge and scientific judgment. AI accelerates execution. Human expertise defines impact. #MassSpectrometry #Proteomics #Bioinformatics #AI #ComputationalBiology

58

8 Comments

Even at temperature zero, LLMs are not deterministic. That should bother us. I have been reading a technical post from Thinking Machines Lab that digs into why. The answer is not what most people assume. The common explanation is that GPUs run computations in parallel, threads finish in random order, and floating-point arithmetic is non-associative—(a + b) + c does not always equal a + (b + c) when precision is finite. True, but incomplete. The deeper issue is “batch invariance.” When the load on an inference server changes, the batch size changes. When the batch size changes, the reduction strategy inside kernels can change. When the reduction strategy changes, the floating-point accumulation order changes. Different order, different result. Same prompt. Same model. Same temperature zero. Eighty different outputs from 1,000 requests. This matters beyond LLMs. Any numerical computation that involves reductions—matrix multiplications, attention mechanisms, but also differential equation solvers, Bayesian samplers, stochastic simulations—faces the same underlying issue. Floating-point arithmetic is deterministic only if you control the accumulation order. And controlling it is harder than it looks. Reproducibility is supposed to be a bedrock of science. We spend enormous effort on statistical rigour, version control, data provenance. But how often do we verify that our numerical code produces bitwise identical results across runs, machines, or batch sizes? The good news: Thinking Machines Lab shows it can be solved. Batch-invariant kernels, consistent reduction strategies, careful engineering. It is not free, but it is tractable. The bad news: most of us have not been thinking about it at all. Link in comments. #MachineLearning #Reproducibility #ComputationalBiology #AIinBiology

97

29 Comments

Over 1.28 lakh candidates qualified for PhD admissions through UGC NET June 2025—a 14% rise from last year. Sounds promising, right? More scholars, more research. But here’s the catch: only 5,269 secured JRF. That’s roughly 4.1%. This growing gap between PhD aspirants and JRF awardees is worrying. Without adequate funding, many students are forced to juggle teaching or part-time work, diluting the very focus and intensity that research demands. Quality research isn’t just about admission numbers. It needs structural support—fellowships, mentorship, and resources that allow scholars to think deeply and innovate freely. If we truly want to strengthen India’s research ecosystem, bridging this funding gap is not optional—it’s urgent. What are your thoughts? Is the bottleneck in funding one of the biggest hurdles for research in India today? #PhD #Research #UGCNET #HigherEducation #FundingGap #JRF #Academia

271

21 Comments

Don't miss our recent collection on Advances in Computational Protein Design, Structural Biology, and Drug Discovery! Guest edited by Aditya Kumar Padhi, Prof. Timir Tripathi and Seetharaman Jayaraman, this collection explores key areas including computational protein and antibody design, vaccine design, enzyme engineering & more. Featuring... "Peptide-based drug design using generative AI" by Srinivasan Ekambaram and Nikolay Dokholyan 👉 https://lnkd.in/ehCET3Pa "Liposome-based delivery of DNA aptamers to inhibit erythromycin methyltransferase-mediated antibiotic resistance" by Swagata Patra, Damini Sahu, Leena Badgujar, P. I. Pradeepkumar and Ruchi Anand from IIT Bombay 👉 https://lnkd.in/eTFas3KZ "Machine learning prediction of multiple distinct high-affinity chemotypes for α-synuclein fibrils" by Xinning Li, Ryann Perez, Zhude Tu, Robert Mach, Sam Giannakoulias and E. James Petersson 👉 https://lnkd.in/eNun97CD All articles in the collection are free to read until the end of April 2026, if not already Open Access 🔓👉 https://lnkd.in/eWyucxbS

41

In a recent study published in the Journal of the American Chemical Society, researchers from IIT Bombay and collaborators from IIT Kanpur and CNCI, Kolkata uncover how collagen accelerates clumping of hormones in the pancreas, identifying a previously unknown diabetes trigger and a potential new drug target. They have identified an important missing link: fibrillar collagen I, a major component of the extracellular matrix. Prof. Shamik Sen from the Department of Biosciences and Bioengineering at IIT Bombay led and oversaw the project. https://lnkd.in/di_b5a3M

190

5 Comments

💯 Absolutely true, Vineet Chaudhary. V Many of us have faced the same issue — we never get to know the real reason behind the rejection. Often, we design our projects based on the available facilities in the host or collaborating institute, and within realistic, achievable objectives for a 2–3 year period. But still, rejections come with vague comments like “Good proposal, but not selected.” There’s no clarity on whether the reviewer found it too basic, lacking scientific merit, or something else entirely. And this isn’t limited to just NPDF — it’s the same story across all funding schemes like DBT, CSIR and others. If a project demonstrates solid scientific merit, it deserves to be funded. At the very least, applicants should be given clear feedback — not left guessing what went wrong. Transparency and constructive evaluation can go a long way in strengthening the research ecosystem, rather than disheartening early-career researchers. #Research #Academia #NPDF #Postdoc #SciencePolicy #Researchers

14

Why do labs with $500,000 sequencing machines still use paper notebooks? As a bench scientist, I've lived this paradox daily - experiencing the frustration of transferring data between disconnected systems, converting file formats, and wrestling with Excel spreadsheets while using cutting-edge equipment. Data transfers, format conversions, figure formatting, protocol hunting - every bench scientist knows this soul-crushing routine. But what if it didn't have to be this way? While everyone talks about AI and automation, a quieter revolution is transforming how scientific research actually gets done. I explored 11 companies tackling the unsexy but critical problem of laboratory data infrastructure, each taking distinct approaches: comprehensive platforms like Benchling, Revvity Signals, Scispot, Sapio Sciences, and Labguru (BioData) that replace multiple lab tools with unified systems; data infrastructure providers such as TetraScience and Ganymede Bio that connect and standardize data across existing tools; and specialized solutions including BenchSci, Collaborative Drug Discovery - CDD Vault, Kaleidoscope.bio, and Synthace that target specific workflow pain points. The market opportunity is massive - every biotech, pharma company, and research institution struggles with disconnected laboratory systems. My deep-dive reveals how these platforms are finally addressing lab data chaos and moving research from data silos to integrated workflows. However, adoption barriers reveal a stark divide between industry capabilities and academic accessibility. Article link in comments. Fellow scientists: What's your biggest lab data pain point? Have you tried any of these platforms? Did I miss any? What would your dream laboratory data management system look like? Drop your thoughts in the comments or do reach out.

20

2 Comments

From B.Tech. Pride to Ph.D. Puzzle: What IITs Don’t Talk About........... About 15 years ago, when I attended an orientation programme at IIT after joining my Ph.D., our Dean of Academics told us: “Our IITs are famous for their undergraduate programs, especially the B.Techs. But when it comes to postgraduate and Ph.D. programs, we still need to improve and match global standards.” Some critics argued even then that IITs were little more than a supply line of graduates for foreign universities and companies. Sadly, not much has changed. The postgraduate and Ph.D. programs still function largely as feeders—sending hardworking postdocs abroad. A few succeed, return, and build careers here; some remain abroad. But for the majority, the Ph.D. leads to an endless loop of “postdoc after postdoc,” with no clear exit. Compared with state or central universities (the so-called “non-privileged”), IITs flaunt how many of their students went abroad and “settled.” Whether that settlement is temporary—shaped by visa restrictions and immigration policies—is another matter. Today, the situation is even worse with growing uncertainties in the West. Yet IITs will never openly admit this. Instead, placement cells glorify undergraduate packages and showcase alumni in the West. But the real question is: do they have any data on their Ph.D. students? Where are they now? How many secured permanent jobs even 10 years after graduation? And how many are simply doing nothing? Most Ph.D. programs (in more than 90% of labs) only generate papers and clones. Graduates are rarely employable in the biotech or chemical industries, barring a few cases where extraordinary PIs defy the system and do Herculean work. Visit any well-known lab’s website and you’ll see pages of publications in JACS, Angewandte Chemie, or Nature Communications. Yet, the CVs of Ph.D. students look indistinguishable—from each other and from their PIs. Those PIs had the “first-mover advantage.” Some of their early students also benefited. But most remain unemployable, lacking the skill sets needed to survive in market-driven industries. If Plan A (an academic job) fails, there is no Plan B. Abroad, even in Nobel Laureates’ labs, only 20–30% of students enter academia, while the rest thrive in industry because they are equipped with real, marketable skills. The lesson is clear: never take career advice from someone who secured a job 20 years ago simply because they spent time at an Ivy League university or co-authored a paper in a high-impact journal. The landscape has changed. We may be world-class in some research areas, but many of these fields are overcrowded, leaving most Ph.D.s unemployable. The only way forward is to diversify skills and remain adaptable to market demands. As the saying goes: “Do not put all your eggs in the same basket.” #PhDLife #IndianAcademia #PostdocLife #SkillDevelopment #BeyondAcademia #ResearchCulture #CareerDevelopment

3