Ichiko Sugiyama

AT50-10 CliOMZ 2023 cruise has been one of the best thing that happened to my life! We get to do awesome science, work with stellar scientists, and see amazing wildlife! Cannot wait to go on another cruise! #WorldOceanDay

We are looking for photos or videos of microscopy images and microbial ecologists at work! The winning photos and videos will be used to promote the ISME Society and of our publications. Submit your entry by 11 November! 🔗isme-microbes.org/call-for-micro… #MicrobialEcology #microbes

After a lecture on principal component analysis in my #MachineLearning course, a student approached me and asked, "But what do the components look like?" That evening, I returned to my office and pondered how I could visually demonstrate to my students how PCA learns and adapts to our data. So, I developed a #Python @matplotlib dashboard that illustrates this process. In it, X1 and X2 are positively correlated, X1 and X3 are negatively correlated, and X4 is independent. You can adjust the strength of these correlations and observe how the loadings and explained variance change! Check it out on my #GitHub repository: github.com/GeostatsGuy/Da… ∀. #DataScience

New online! Models translate lake mud into useful climate variables bit.ly/4gVmxdD

This is how you can assign colors in #tidyplots 🌈 #rstats #dataviz #phd

Two weeks left for students and postdocs to apply to receive support for international research, training, or lab visits. Submit your application by 1 October 2025! Learn more and apply: 🔗isme-microbes.org/isme-scholar-m… #MicrobialEcology #microbes

Truth 😅

📣#MTTM Fadime Stemmer is a MIT/@WHOI Joint Program PhD Student in Chemical Oceanography working with @MakASaito. She is investigating the role that proteases (enzymes that degrade proteins) play in marine carbon cycling & how trace metal availability impacts these processes.

Sturtian Barite! pubs.geoscienceworld.org/gsa/geology/ar…

Congratulations Jonathan, Hanika and the team! Evidence for Hadean mafic intrusions in the Nuvvuagittuq Greenstone Belt, Canada | Science science.org/doi/10.1126/sc… #Carleton #UOttawa

📣 Interested in meeting new collaborators and projecting changes in ocean metabolisms 🌊over different space & time scales using modeling? Sign up for the BioGeoSCAPES Modeling Workshop, co-sponsored by C-CoMP, @WHOI in September 2025. Apply by 6/30: tinyurl.com/2ub5zx8p.

ASM’s Agar Art Exhibit is live tonight at E2Art Gallery in LA! Guests are exploring artwork created with living microbes by scientists and artists from around the world. Thanks for joining us to celebrate where art meets science. View the collection: asm.social/2tc

💥New Paper! Depositional Controls on ∆′17O Signatures of Sedimentary Sulfate - Crockford - 2025 - Geophysical Research Letters - Wiley Online Library agupubs.onlinelibrary.wiley.com/doi/10.1029/20… Thanks to all of the folks who helped this over the finish line.

Nature research paper: Diet outperforms microbial transplant to drive microbiome recovery in mice. go.nature.com/3YXbf0i

Principal Component Analysis (PCA) is a powerful tool that helps reduce the dimensionality of large datasets while retaining the most relevant information. pyoflife.com/principal-comp… #DataScience #rstats #dataanalysis #statistics #mathematics #MachineLearning #datascientists

In a new Science study, researchers estimate that as many as 1.4 billion people live in areas with soil dangerously polluted by heavy metals like arsenic, cadmium, cobalt, chromium, copper, nickel, and lead. The results reveal a global risk, but also a previously unrecognized high-risk, metal-enriched zone in low-latitude Eurasia. scim.ag/4jdqOJB

This book explores practical steps and considerations for leveraging machine learning in data analysis using Python. pyoflife.com/practical-mach… #DataScience #machinelearning #ArtificialInteligence #datavisualizations #statistics #mathematics

Trying to choose the right clustering method for your data? K-means and Expectation-Maximization (EM) clustering are two popular options, but they rely on different assumptions and can lead to very different results depending on the structure of your data set. Here are some guidelines to help you choose the method that fits your data best. ✔️ K-means is fast, easy to apply, and works well when clusters are compact, spherical, and similar in size ✔️ EM, often used with Gaussian Mixture Models, handles clusters with different shapes, sizes, and orientations ✔️ EM provides soft assignments, which means each point is assigned a probability of belonging to each cluster ❌ K-means forces hard assignments and assumes equal variance, which can distort results on uneven structures ❌ EM is more computationally demanding and can overfit if the number of components isn’t chosen carefully ❌ Both methods are sensitive to initialization and can converge to suboptimal solutions To improve outcomes, consider scaling your data, using PCA before clustering, and selecting the number of clusters using criteria like BIC, AIC, or silhouette score. Running multiple initializations can also help avoid poor results. The image compares both methods on a data set called "mouse." K-means splits the space into equal partitions, missing the natural structure. EM fits flexible Gaussian distributions that better match the true cluster shapes. Credit for the visualization: en.wikipedia.org/wiki/K-means_c… 🔹 In R, use stats::kmeans() for K-means and mclust::Mclust() for EM, which includes built-in tools for model comparison and visualization. 🔹 In Python, use scikit-learn with KMeans and GaussianMixture, and apply PCA or model selection metrics like bic and silhouette_score. For regular insights on practical statistics, data analysis, and how to apply them in R and Python, join my newsletter. Click this link for detailed information: eepurl.com/gH6myT #R #Statistics #DataAnalytics #statisticians #RStats #datastructure

Today in my #MachineLearning course, we're diving into Convolutional Neural Networks (CNNs) 🧠📊 — a powerful technique in image-based #DeepLearning 🤖. To keep it practical, I’ve built a simple example that classifies random vs. smooth images 🎨📷. We'll train the model, evaluate its performance 📈, and even visualize the learned kernels and feature maps 🔍🧩 to see what the network is learning! I share it on #GitHub @ github.com/GeostatsGuy/Py… ∀.

Making a heatmap is an essential skill for a bioinformatician. But you probably do not understand heatmap. 7 reading resources to understand heatmap! 🧵

There are many applications of this work, both modern and ancient surface environments, which can be found in the discussion section 😉!

Paper alert 🚨 Interaction of metal oxyions with carbonate green rust in seawater is out now! This is one of the few papers that investigated metal uptake with GRCO3 in seawater and conducted 1 year aging experiments! Link here: sciencedirect.com/science/articl… #biogeochemistry #metals